[{"data":1,"prerenderedAt":2703},["ShallowReactive",2],{"presentations-en-webinar_wissensmanagement":3,"all-presentations-pages-en":153},{"id":4,"title":5,"audience":6,"body":7,"carouselItems":136,"companyName":136,"date":136,"description":137,"end_date":136,"eventDate":138,"eventDuration":139,"eventName":140,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":142,"mainTitle":136,"meta":143,"navigation":144,"path":145,"podcastId":136,"role":136,"seo":146,"series":136,"start_date":136,"stats":136,"stem":147,"subtitle":148,"tags":136,"techs":136,"titleHighlight":149,"videoId":150,"views":151,"watchTime":136,"__hash__":152},"presentations\u002Fpresentations\u002Fen\u002Fwebinar_wissensmanagement.md","Webinar: Knowledge Management","ca. 70 participants",{"type":8,"value":9,"toc":123},"minimark",[10,14,18,23,26,32,35,39,42,45,49,52,57,60,64,67,70,74,77,80,83,88,92,95,99,102,105,110,113,117,120],[11,12,13],"p",{},"On 3 January 2025 bbv held the second episode of its webinar series on artificial intelligence. The event was moderated by Stefan Herberling and Alan Ettlin, who together with me offered in-depth insights into modern knowledge management using AI agents. The goal of the webinar was to look beyond the current hype about artificial intelligence and to illuminate the topic from both a business and a technical perspective.",[15,16],"youtube-video",{":video-id":17},"videoId",[19,20,22],"h3",{"id":21},"knowledge-management-in-focus","Knowledge Management in Focus",[11,24,25],{},"At the start of the webinar I introduced the DIKW pyramid, a fundamental structure running from data through information and knowledge to wisdom. This pyramid serves as a theoretical framework for understanding the different levels of knowledge management and designing them effectively.",[27,28,29],"blockquote",{},[11,30,31],{},"\"Knowledge management is not just the administration of databases; it encompasses processes, methods and practices for the shared use and management of information.\"",[11,33,34],{},"This statement underscored the importance of efficient knowledge management as a decisive success factor for achieving organisational goals. I explained that knowledge management goes far beyond the simple administration of data. It encompasses the shared use and management of information that is necessary to achieve organisational objectives. This is not only about technical systems but also about the processes and practices that ensure knowledge is effectively shared and used.",[19,36,38],{"id":37},"definition-and-context-of-knowledge-management","Definition and Context of Knowledge Management",[11,40,41],{},"Knowledge management is a complex field that encompasses both strategic and operational aspects. In my presentation I emphasised that it is not merely about storing and managing data but rather about fostering a culture of knowledge sharing and continuous improvement. \"In modern organisations we have increasingly become knowledge-based enterprises. Effective knowledge management is therefore indispensable for remaining competitive and driving innovation.\"",[11,43,44],{},"I explained further that a central goal of knowledge management is not only to collect knowledge but also to make it accessible and usable. This requires a combination of technological solutions and organisational measures that ensure knowledge is effectively shared and used. Both explicit knowledge, which can easily be documented and shared, and implicit knowledge, which often resides in employees' heads, play an important role.",[19,46,48],{"id":47},"live-demonstration-ai-agents-in-action","Live Demonstration: AI Agents in Action",[11,50,51],{},"The highlight of the webinar was the live demo in which I showed how AI agents can proactively support knowledge management. Using a fictitious company called \"Eisbeit\" operating in food production, I demonstrated how new employees can quickly access company-specific knowledge through interaction with AI agents.",[27,53,54],{},[11,55,56],{},"\"Examining information from different perspectives and thereby generating wisdom,\" I explained during the demo, which significantly increased the efficiency and accuracy of knowledge transfer.",[11,58,59],{},"In the demonstration it became clear how AI agents not only make explicit knowledge accessible but can also actively promote and integrate implicit knowledge. For example, a new employee could obtain information about internal processes and specific company terms through a simple query to the AI agent, without needing to complete extensive training. The AI agents were able to bring together data from different departments, contextualise it and put it into usable form. This means knowledge is not only centralised but also provided dynamically and interactively.",[19,61,63],{"id":62},"application-examples-from-practice","Application Examples from Practice",[11,65,66],{},"During the demo I showed concrete examples of how AI agents can be used in a company like \"Eisbeit\". One example was the management of company-specific terms such as \"Frostflow\". Through the integration of a glossary, the AI agent could automatically recognise and explain these terms, which is of great benefit especially for new employees. This significantly reduced onboarding time and promoted efficiency in the workplace.",[11,68,69],{},"A further example was the monitoring and analysis of production data. The AI agent could monitor temperature data in real time and immediately raise an alarm on deviations in order to ensure product quality. This impressively showed how AI agents can not only respond passively to requests but actively contribute to improving business processes.",[19,71,73],{"id":72},"future-perspectives-and-roadmap","Future Perspectives and Roadmap",[11,75,76],{},"After the live demonstration I presented a comprehensive roadmap for implementing AI agents in knowledge management. The first step is to start with simple language models like Fast GPT and Smart GPT. These models form the foundation and enable rapid introduction without immediate adaptations.",[11,78,79],{},"Building on this, the integration of proprietary knowledge databases should follow to provide internal company knowledge. In the third step, proactive agents could be introduced that actively ask follow-up questions and continuously expand the knowledge database. This means AI agents not only respond to requests but can also proactively gather information and expand knowledge within the organisation.",[11,81,82],{},"Finally, additional information channels could be monitored and the agents individually tailored to employee preferences. This includes adapting communication styles and integrating personal working habits to make interaction with the AI agents as effective as possible.",[27,84,85],{},[11,86,87],{},"\"There are many ways we can further improve these agents,\" I summarised, emphasising the flexibility and scalability of the proposed solutions.",[19,89,91],{"id":90},"long-term-visions","Long-Term Visions",[11,93,94],{},"The roadmap closes with long-term visions in which AI agents function as an integral part of the organisational structure. This includes the continuous further development of agents through machine learning and the integration of new technologies in order to always remain at the cutting edge of development. The goal is to create an intelligent and adaptive knowledge management infrastructure that flexibly adapts to the changing needs of the organisation.",[19,96,98],{"id":97},"interactive-qa","Interactive Q&A",[11,100,101],{},"In the subsequent Q&A section numerous interesting questions from the audience were addressed. Topics at the centre included automated creation of glossaries, the four-eyes principle for expert knowledge, security and data protection when integrating ERP systems, and the lifecycle management of content.",[11,103,104],{},"On the question of how to avoid incorrect knowledge I explained:",[27,106,107],{},[11,108,109],{},"\"Algorithms can be used to detect contradictions in the data and take appropriate measures.\"",[11,111,112],{},"I emphasised the importance of automated processes combined with human verification to ensure the quality and reliability of knowledge management. Further questions addressed the integration of data from ERP systems and the possibility of developing AI agents to automatically anonymise sensitive data. Particularly noteworthy was the discussion on data security and data protection, where I explained how important it is to choose trustworthy providers and establish clear guidelines for handling sensitive information.",[19,114,116],{"id":115},"challenges-and-approaches","Challenges and Approaches",[11,118,119],{},"A central topic was the integration of AI agents into existing IT infrastructures. It was emphasised that careful planning and step-by-step implementation are necessary to ensure compatibility with existing systems. I explained various approaches to data integration and underscored the importance of interfaces that enable seamless communication between AI agents and existing IT systems.",[11,121,122],{},"Another important topic was the scalability of AI solutions. I explained that scalability plays a central role in meeting the rising requirements of a growing organisation. Through the use of modular architectures and flexible cloud solutions, organisations can ensure that their AI agents grow with the organisation and adapt to changing requirements.",{"title":124,"searchDepth":125,"depth":125,"links":126},"",2,[127,129,130,131,132,133,134,135],{"id":21,"depth":128,"text":22},3,{"id":37,"depth":128,"text":38},{"id":47,"depth":128,"text":48},{"id":62,"depth":128,"text":63},{"id":72,"depth":128,"text":73},{"id":90,"depth":128,"text":91},{"id":97,"depth":128,"text":98},{"id":115,"depth":128,"text":116},null,"AI agents are revolutionising work processes and increasing team efficiency. OpenAI impressively demonstrated\nwith Custom GPTs what extraordinary synergies emerge when several of these intelligent assistants work\ntogether. The potential for transforming your organisation is enormous. We explain how generative AI can\nchange your organisation for the better and what steps are needed to deploy AI agents successfully. We show\npractically, using concrete examples, how a multi-agent system can be actively used, and we also venture a\nlook into the future and what it holds for AI agents.\n","2024-01-10","63 Min.","bbv KI Webinar - Knowledge Management","md","\u002Fimages\u002Fpresentations\u002Fwebinar2\u002Fwissensmanagement.png",{},true,"\u002Fpresentations\u002Fen\u002Fwebinar_wissensmanagement",{"title":5,"description":137},"presentations\u002Fen\u002Fwebinar_wissensmanagement","Expert Knowledge on Demand with AI Agents","Knowledge Management","ZvtoqaiCijI","850","2keJLgWHFbeyX7YspEyC6dgNGcZZGCUyeedfLJUljU8",[154,220,275,380,512,629,724,830,929,1055,1211,1336,1560,1751,1905,2039,2202,2331,2513,2618],{"id":155,"title":156,"audience":157,"body":158,"carouselItems":205,"companyName":136,"date":136,"description":208,"end_date":136,"eventDate":209,"eventDuration":210,"eventName":211,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":212,"mainTitle":136,"meta":213,"navigation":144,"path":214,"podcastId":136,"role":136,"seo":215,"series":136,"start_date":136,"stats":136,"stem":216,"subtitle":217,"tags":136,"techs":136,"titleHighlight":218,"videoId":136,"views":136,"watchTime":136,"__hash__":219},"presentations\u002Fpresentations\u002Fen\u002F9ter_Datenschutztag.md","Data Flows in AI Systems","40-50 participants",{"type":8,"value":159,"toc":199},[160,163,167,170,174,177,181,184,188,191,194],[11,161,162],{},"On 15 March 2024 I had the privilege of giving a talk on data flows in AI systems at the 9th Datenschutztag AI & Data Protection organised by HÄRTING Rechtsanwälte AG.\nThe event was aimed primarily at professionals from the fields of data protection and information security who face the growing challenges around the use of artificial intelligence in their organisations. In my talk I shed light on fundamental questions that arise in connection with generative AI systems, particularly around the capture, processing, and sharing of data. With my presentation I aimed to give a better understanding of the complex data movements in AI systems and to show how companies can manage these processes safely and efficiently. My intention was to lay the groundwork for a well-informed discussion, especially in combination with the other more legally oriented talks of the day.\nAt the same time, global legislative initiatives, such as the EU AI Act, were a central theme of the event, defining the prerequisites for safe, transparent, and responsible AI use.",[19,164,166],{"id":165},"generative-ai-and-the-role-of-llms","Generative AI and the Role of LLMs",[11,168,169],{},"In the first section of my presentation I explained how Large Language Models (LLMs) work: they learn from enormous datasets using statistical methods, recognising patterns in order to independently generate text, images, or other content. I emphasised that this so-called \"generative\" potential forms the basis for many innovative applications but also creates new risks. Data leaks can occur when sensitive information in training or prompt data is inadequately protected. This is a core challenge: with every new prompt and every interaction, confidential knowledge can be inadvertently disclosed. I also stressed that this risk does not originate directly from LLM technology itself but is a general problem of data movements in which third-party services are used.",[19,171,173],{"id":172},"context-and-prompt-structure","Context and Prompt Structure",[11,175,176],{},"A further focus was on how important it is to \"feed\" the AI with the right information. Prompting refers to steering LLMs through targeted input that must be embedded in a specific context. Using practical examples, I showed how context plays a decisive role in the quality and relevance of AI responses. The Retrieval Augmented Generation (RAG) method illustrates this particularly well: external data sources are integrated and searched in a structured way using vector databases to provide suitable input for the prompt. The more precisely this context is prepared and the more skilfully prompts are formulated, the more accurately and safely the AI system can act. However, companies must ensure that no sensitive information is transmitted uncontrolled to third-party services in this process.",[19,178,180],{"id":179},"challenges-in-handling-data","Challenges in Handling Data",[11,182,183],{},"In the third section I explained the hurdles companies frequently face when deploying AI systems. These include above all the limited context lengths of some generative models, which make intelligent management and prioritisation of the data provided necessary. Thorough validation of AI outputs is also indispensable, since LLMs are excellent at recognising and replicating patterns but do not have genuine factual knowledge. This can result in so-called \"hallucinations\": invented or incomplete information. The situation becomes even more sensitive when companies rely on confidential data that must be protected or anonymised. Using concrete case studies I illustrated how data sources can be designed to guarantee the highest possible level of security. Anonymisation and pseudonymisation concepts are indispensable here to comply with legal data protection requirements while still benefiting from the enormous potential of AI.",[19,185,187],{"id":186},"discussion-and-outlook","Discussion and Outlook",[11,189,190],{},"The subsequent discussion round made clear that the legal framework, not only in Europe but also in the USA and other regions, strongly influences how companies deploy AI technologies. Data Protection Officers (DPO) and Chief Information Security Officers (CISO) play a key role in risk analysis, supplier management, and the protection of critical infrastructure. Many audience members shared their experiences with AI solutions and brought valuable questions ranging from technical details about vector databases to ethical considerations on the use of AI.",[11,192,193],{},"To conclude: the future of artificial intelligence depends not only on technical progress but significantly on responsible data management. Anyone who wants to use AI efficiently and safely must engage deeply with data flows in AI systems and develop an awareness of possible risks. At the same time the technology holds enormous potential for innovation and value creation, provided all stakeholders, from developers through data protection experts to executives, work closely together and have clear guidelines at hand. These insights are what participants take away from the 9th Datenschutztag AI & Data Protection, and I hope they will implement them in their own organisations to fully harness the opportunities of AI while preserving data protection.",[195,196],"image-carousel",{":height":197,":items":198,":width":197},"400","carouselItems",{"title":124,"searchDepth":125,"depth":125,"links":200},[201,202,203,204],{"id":165,"depth":128,"text":166},{"id":172,"depth":128,"text":173},{"id":179,"depth":128,"text":180},{"id":186,"depth":128,"text":187},[206,207],"\u002Fimages\u002Fpresentations\u002Fhaerting\u002F1.png","\u002Fimages\u002Fpresentations\u002Fhaerting\u002F2.png","Data flows in AI systems are complex and multifaceted. The strategic arrangement of system components in which\ndata is captured, stored, and transmitted forms the foundation of our analysis. By carefully examining the\ndynamics and interactions in typical generative AI systems, we aim to identify and raise awareness of both\ndirectly visible and concealed data streams. Our primary goal is ultimately to sharpen awareness of potential\ndata leaks in order to support and promote the stability, sustainability, and integrity of AI systems.\n","2024-03-15","ca 20 Min.","9th Datenschutztag AI & Data Protection","\u002Fimages\u002Fpresentations\u002Fhaerting\u002Fhaerting_3.png",{},"\u002Fpresentations\u002Fen\u002F9ter_datenschutztag",{"title":156,"description":208},"presentations\u002Fen\u002F9ter_Datenschutztag","Understanding Data Movements and Their Implications","AI & Data Protection","H3tYqHk3dXyQa1wbdKUQRb-UzvixlT1QJPy0h1qNPwE",{"id":221,"title":222,"audience":223,"body":224,"carouselItems":136,"companyName":136,"date":136,"description":263,"end_date":136,"eventDate":264,"eventDuration":210,"eventName":265,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":266,"mainTitle":136,"meta":267,"navigation":144,"path":269,"podcastId":136,"role":136,"seo":270,"series":136,"start_date":136,"stats":136,"stem":271,"subtitle":272,"tags":136,"techs":136,"titleHighlight":273,"videoId":136,"views":136,"watchTime":136,"__hash__":274},"presentations\u002Fpresentations\u002Fen\u002Fbbv_ai_impact.md","AI in Action at FMH","100-150 participants",{"type":8,"value":225,"toc":257},[226,229,233,236,240,243,247,250,254],[11,227,228],{},"On 28 August 2024 I co-presented a panel discussion with Ms Stephanie Wyler-Marti of FMH under the title \"AI in Action at FMH: AI as the Key to a Successful Tariff Structure Transition from Tarmed to Tardoc\". FMH, the Foederatio Medicorum Helveticorum, is the umbrella organisation of Swiss physicians and brings together more than 70 medical organisations with over 45,000 individual members. In this capacity it carries out central functions including tariff development, quality assurance, and continuing medical education. The transition from Tarmed to Tardoc means a new nomenclature, new rules, and the introduction of additional chapters and tariffs for all outpatient services. This creates an enormous need for information and training for medical practices and for FMH itself. The joint discussion aimed to examine how modern AI technologies can help address these new challenges efficiently.",[19,230,232],{"id":231},"the-jump-from-tarmed-to-tardoc","The Jump from TARMED to TARDOC",[11,234,235],{},"The switch from Tarmed to Tardoc represents a major step for the entire Swiss healthcare system. Numerous new services, split billing, and a more complex structure increase the workload for all involved. Practice owners need to adapt their billing systems, train staff, and ensure that all new requirements are correctly implemented. At the same time FMH faces the task of providing competent information about the tariff change, answering individual questions, and identifying tariff gaps quickly. The transition affects every outpatient physician in Switzerland, so the volume of enquiries is correspondingly high.",[19,237,239],{"id":238},"why-ai","Why AI?",[11,241,242],{},"In my presentation I explained the reasons why FMH is seriously considering AI in the context of the tariff change. On one hand there are modern options such as chatbots or GPT solutions that are ideally suited to answering frequently recurring questions. On the other hand an AI-assisted solution offers the opportunity to prepare FMH's accumulated tariff knowledge in such a way that many requests can be handled in an automated yet high-quality manner. Such an approach not only improves service quality but also relieves staff, allowing them to focus on more complex requests and individual consultations. Not least, FMH gained experience with a high volume of enquiries as far back as 2018, which motivated them to explore new technologies to increase member satisfaction.",[19,244,246],{"id":245},"technical-and-organisational-challenges","Technical and Organisational Challenges",[11,248,249],{},"The presentation addressed in detail how an AI solution could look that accesses only internal information and does not \"invent\" incorrect answers. This is a core challenge: the AI must be fed from a reliable database and may only reproduce content that has been validated beforehand. An agent concept is suited to this, in which specialised agents draw on different data sources depending on the question. For example, a \"handbook agent\" could provide general information from the FMH handbook, while a \"rules agent\" has access to position-specific billing rules. Technical restrictions would also be needed to prevent the AI from drawing on unsecured information from external sources. An agent log that records all questions and answers is useful for tracing how an answer was reached and whether knowledge gaps exist. This also allows potential error sources to be quickly identified and the knowledge base to be continually improved.",[19,251,253],{"id":252},"experiences-and-outlook","Experiences and Outlook",[11,255,256],{},"At the end of the presentation I outlined the possible impact of AI use on FMH's working practices. The relief provided by automated responses would free up capacity for demanding and more personal consultations. At the same time FMH could draw valuable insights from the enquiries to identify tariff gaps and close them more quickly if needed. AI technology can also be integrated into further areas of FMH over time, from continuing education to quality assurance. The tariff change has shown that the healthcare sector faces increasingly complex tasks that are difficult to manage with traditional methods. AI offers a modern approach to making processes more efficient and meeting the high expectations of members. The potential remains large going forward: ongoing optimisation and responsible data handling can help to support the tariff transition sustainably and at the same time pave the way for further digital innovation.",{"title":124,"searchDepth":125,"depth":125,"links":258},[259,260,261,262],{"id":231,"depth":128,"text":232},{"id":238,"depth":128,"text":239},{"id":245,"depth":128,"text":246},{"id":252,"depth":128,"text":253},"The transition from Tarmed to Tardoc and the associated challenges for Swiss physicians are the central focus.\nThe discussion examines how FMH can use AI technologies, including chatbots and GPT solutions, to support\ninformation and advisory work. Organisational and technical aspects are addressed alongside a possible future\nperspective for secure and progressive AI use.\n","2024-08-28","bbv - AI Impact Forum","\u002Fimages\u002Fpresentations\u002Fai_impact\u002Fai_impact_2.png",{"links":268},"https:\u002F\u002Fbbv.ch\u002Fimpressionen-swiss-ai-impact-forum-2024\u002F","\u002Fpresentations\u002Fen\u002Fbbv_ai_impact",{"title":222,"description":263},"presentations\u002Fen\u002Fbbv_ai_impact","Successful Tariff Structure Transition from TARMED to TARDOC","AI Impact Forum","GR6-hnDFqLLmKKDvyOlHKAiTxzxbYO4XOstpanc8Rpk",{"id":276,"title":277,"audience":278,"body":279,"carouselItems":361,"companyName":136,"date":136,"description":368,"end_date":136,"eventDate":369,"eventDuration":370,"eventName":371,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":372,"mainTitle":136,"meta":373,"navigation":144,"path":374,"podcastId":136,"role":136,"seo":375,"series":136,"start_date":136,"stats":136,"stem":376,"subtitle":377,"tags":136,"techs":136,"titleHighlight":378,"videoId":136,"views":136,"watchTime":136,"__hash__":379},"presentations\u002Fpresentations\u002Fen\u002Fbbv_ki_revolution.md","Application Areas","30-40 participants",{"type":8,"value":280,"toc":351},[281,285,288,292,295,299,302,306,309,313,316,320,324,327,331,334,338],[19,282,284],{"id":283},"a-look-at-the-future-of-ai-recap-of-the-ai-revolution-event","A Look at the Future of AI: Recap of the \"AI Revolution\" Event",[11,286,287],{},"At the \"AI Revolution\" event, hosted by bbv Software Services AG, we had the opportunity to introduce our clients to the fascinating possibilities of artificial intelligence. Under the heading AI Revolution, the new and rapidly growing technology of generative AI was the focus. The aim of the event was to make the potential of modern AI technologies tangible and accessible.",[19,289,291],{"id":290},"inspiring-welcome-and-introduction","Inspiring Welcome and Introduction",[11,293,294],{},"The event opened with a warm welcome from Stefan Häberling and Alan Ettlin.\nThey addressed inspiring words to participants and created a motivating atmosphere for the content that followed. Immediately afterwards Joel Barmettler presented his talk on AI agents. He explained not only how these work but also how they act and in what different forms they can appear. Particularly noteworthy was his explanation of the adoption levels of AI agents, ranging from simple to highly complex systems.",[19,296,298],{"id":297},"technical-foundations-and-deeper-insights","Technical Foundations and Deeper Insights",[11,300,301],{},"Later in the event Cedric Klinkert introduced participants to the technical foundations of modern AI technologies. His presentation shed light on how language models work at their core and what mechanisms underlie concepts such as Retrieval Augmented Generation (RAG). With clear explanations and practical examples he made these complex technologies accessible and understandable for the audience.",[19,303,305],{"id":304},"practical-applications-in-the-bbv-ai-hub","Practical Applications in the BBV AI Hub",[11,307,308],{},"The centre of my own presentation was the BBV AI Hub, our platform for developing and testing AI solutions. I demonstrated various AI agents in action to illustrate the wide range of possible use cases. Particularly impressive was the so-called LinkedIn agent, which is able to create social media posts in line with our company guidelines. Another example showed how a RAG agent works with our internal company wiki to respond to employee questions in a targeted way and prepare relevant information. I also presented an interactive group chat in which multiple AI agents interact with one another to jointly solve complex problems. These practical demonstrations made it clear how AI technologies can simplify and optimise workflows.",[19,310,312],{"id":311},"generative-ai-and-its-impact-on-companies","Generative AI and Its Impact on Companies",[11,314,315],{},"Alan Ettling then focused on the effects of generative AI in a business context. In his presentation he examined how companies can work more efficiently and at the same time become more innovative through the use of such technologies. He also addressed the strategic challenges and opportunities that arise from integrating AI solutions.",[195,317],{":height":318,":items":198,":width":319},"450","600",[19,321,323],{"id":322},"interactive-closing-with-panel-discussion","Interactive Closing with Panel Discussion",[11,325,326],{},"At the close of the event all speakers came together for a panel discussion.\nIn this interactive format participants had the opportunity to ask open questions and gain deeper insights into the topics presented. The lively discussions showed how great the interest in and enthusiasm for the AI solutions on display were.",[19,328,330],{"id":329},"an-inspiring-day-full-of-new-perspectives","An Inspiring Day Full of New Perspectives",[11,332,333],{},"The \"AI Revolution\" event was not only a showcase for modern AI technologies but also a place for exchanging ideas and visions. Participants left the event with new impulses and a clearer understanding of how artificial intelligence can enrich their work processes and business models. This journey into the world of AI once again showed how close the future already is.",[19,335,337],{"id":336},"links","Links",[339,340,341],"ul",{},[342,343,344],"li",{},[345,346,350],"a",{"href":347,"rel":348},"https:\u002F\u002Fwww.linkedin.com\u002Ffeed\u002Fupdate\u002Furn:li:activity:7113196433511510016\u002F",[349],"nofollow","LinkedIn post by bbv about the event",{"title":124,"searchDepth":125,"depth":125,"links":352},[353,354,355,356,357,358,359,360],{"id":283,"depth":128,"text":284},{"id":290,"depth":128,"text":291},{"id":297,"depth":128,"text":298},{"id":304,"depth":128,"text":305},{"id":311,"depth":128,"text":312},{"id":322,"depth":128,"text":323},{"id":329,"depth":128,"text":330},{"id":336,"depth":128,"text":337},[362,363,364,365,366,367],"\u002Fimages\u002Fpresentations\u002Fki_revolution\u002Fimg_1.png","\u002Fimages\u002Fpresentations\u002Fki_revolution\u002Fimg_2.png","\u002Fimages\u002Fpresentations\u002Fki_revolution\u002Fimg_3.png","\u002Fimages\u002Fpresentations\u002Fki_revolution\u002Fimg_4.png","\u002Fimages\u002Fpresentations\u002Fki_revolution\u002Fimg_5.png","\u002Fimages\u002Fpresentations\u002Fki_revolution\u002Fimg_6.png","At the \"AI Revolution\" event, bbv Software Services AG demonstrated how generative AI technologies can transform\nwork processes. From hands-on demonstrations of innovative AI agents to strategic insights into business\napplications, the event offered an inspiring look at the future of artificial intelligence. Interactive exchange\nwith participants and experts rounded off the day and left plenty of new impulses.\n","2023-09-27","ca 15 Min.","bbv - AI Revolution","\u002Fimages\u002Fpresentations\u002Fki_revolution\u002Fki_revolution.png",{"links":347},"\u002Fpresentations\u002Fen\u002Fbbv_ki_revolution",{"title":277,"description":368},"presentations\u002Fen\u002Fbbv_ki_revolution","Practical Examples of AI Agents","AI Revolution","RPDql68M_FcsF4K39ugv8hVp4fJkZs0hC6iIgiOQeYk",{"id":381,"title":382,"audience":383,"body":384,"carouselItems":136,"companyName":136,"date":136,"description":498,"end_date":136,"eventDate":499,"eventDuration":500,"eventName":501,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":502,"mainTitle":136,"meta":503,"navigation":144,"path":504,"podcastId":136,"role":136,"seo":505,"series":136,"start_date":136,"stats":136,"stem":506,"subtitle":507,"tags":136,"techs":136,"titleHighlight":508,"videoId":509,"views":510,"watchTime":136,"__hash__":511},"presentations\u002Fpresentations\u002Fen\u002Fcosmos-db-conference.md","Orchestrating Intelligent Agents","ca. 2000 live participants",{"type":8,"value":385,"toc":490},[386,389,392,396,399,404,407,411,414,419,422,426,429,434,437,441,444,449,452,456,459,464,467,471,474,479,482,487],[11,387,388],{},"At the Azure Cosmos DB Conference 2025, organised by Microsoft, I had the opportunity to share some key insights from our work with AI agents in a short lightning talk. In just five minutes I showed how we built a system that allows multiple specialised AI agents to work together efficiently, with the help of Azure Cosmos DB.",[15,390],{":video-id":17,":webp":391},"true",[19,393,395],{"id":394},"the-idea-not-one-large-agent-but-many-small-ones","The Idea: Not One Large Agent but Many Small Ones",[11,397,398],{},"For over two years we have been developing at bbv Software Services AG in Zurich a system based on the collaboration of multiple specialised AI agents. The central idea: instead of delegating all tasks to a single generalist model, we distribute them across specialised agents, each responsible for a specific sub-aspect.",[27,400,401],{},[11,402,403],{},"\"An agent with specific knowledge is easier to steer, drifts less from context, and is therefore considerably more reliable.\"",[11,405,406],{},"This division makes the overall system more robust and also allows complex tasks to be handled more efficiently. However, coordinating multiple agents also brings new challenges.",[19,408,410],{"id":409},"communication-between-agents-more-than-just-text","Communication Between Agents: More Than Just Text",[11,412,413],{},"At first it seemed natural to handle communication between agents simply via text, since language models are trained on it. But it quickly became clear that much more structure is needed.",[27,415,416],{},[11,417,418],{},"\"An agent needs to know what the other agents can do, only then can it interact meaningfully.\"",[11,420,421],{},"To enable this interaction we drew inspiration, among other things, from the Model Context Protocol (MCP) proposed by Anthropic. This protocol distinguishes between clients (e.g. LLMs) and servers that provide certain information or execute actions. These servers expose so-called tools: clearly described functions with defined inputs and expected outputs.",[19,423,425],{"id":424},"agents-as-clients-and-servers","Agents as Clients and Servers",[11,427,428],{},"What does this have to do with agents? Simply put: an agent can act in the role of both client and server. An example from our system: an agent is tasked with providing a weather report. Instead of possessing the knowledge itself, it can call on another agent that offers a \"weather API\".",[27,430,431],{},[11,432,433],{},"\"By publishing tool definitions, agents become mutually accessible: they learn to build on each other.\"",[11,435,436],{},"This modularisation makes it possible to build a network of agents that can be extended flexibly. Not every task can be solved with a single API call, though; more complex tasks require coordination and progress tracking.",[19,438,440],{"id":439},"structuring-tasks-runs-threads-and-states","Structuring Tasks: Runs, Threads, and States",[11,442,443],{},"To structure complex workflows we use concepts such as Runs and Threads, as described in LangChain. A Thread bundles all steps of a larger undertaking (e.g. booking a holiday), while a Run represents a single action within that context (e.g. booking a flight).",[27,445,446],{},[11,447,448],{},"\"Complex tasks require an internal state model: that was a central aspect of our agent orchestration.\"",[11,450,451],{},"This is where Azure Cosmos DB comes in: we use the database to reliably store the state of these workflows. Both Runs and Threads are stored in Cosmos DB, as are the tool definitions of our agents.",[19,453,455],{"id":454},"why-cosmos-db","Why Cosmos DB?",[11,457,458],{},"Cosmos DB gives us exactly the flexibility and speed our system requires: a schema-less data model, consistent latency, and global availability.",[27,460,461],{},[11,462,463],{},"\"The flexible structure and real-time synchronisation of Cosmos DB allow us to keep agent states current worldwide.\"",[11,465,466],{},"Particularly in today's dynamic AI landscape it is essential to be able to respond quickly to new developments. Cosmos DB supports us in testing new protocols, changing tool definitions, or improving the internal logic of agents, without costly migrations.",[19,468,470],{"id":469},"agents-beyond-company-boundaries","Agents Beyond Company Boundaries",[11,472,473],{},"To close my talk I broadened the perspective. Most companies are currently building agents only for internal use. But I am convinced that the real strength lies in collaboration across company boundaries.",[27,475,476],{},[11,477,478],{},"\"I envision a world in which agents are rented out, integrated via APIs, negotiate contracts, and jointly implement projects.\"",[11,480,481],{},"This requires common protocols and systems ready to connect with the outside world. That is why we closely follow the development of protocols like the Model Context Protocol. With the recent announcement by OpenAI to support this protocol in future, we see a clear direction.",[27,483,484],{},[11,485,486],{},"\"We continuously adapt our system to new standards, with Cosmos DB as the central platform for agents, tools, and states.\"",[11,488,489],{},"This talk was an opportunity for me to share our experience and perhaps inspire other developers to think about their agent systems more in terms of modularity and collaboration.",{"title":124,"searchDepth":125,"depth":125,"links":491},[492,493,494,495,496,497],{"id":394,"depth":128,"text":395},{"id":409,"depth":128,"text":410},{"id":424,"depth":128,"text":425},{"id":439,"depth":128,"text":440},{"id":454,"depth":128,"text":455},{"id":469,"depth":128,"text":470},"In this lightning talk you will learn how Azure Cosmos DB serves as the foundation for scalable multi-agent\nsystems and enables seamless communication and collaboration between AI agents. Explore practical implementations\nof AI agent protocols within Azure Cosmos DB.\n","2025-04-15","5 Min.","Azure Cosmos DB Conference 2025","\u002Fimages\u002Fpresentations\u002Fcosmos_conf\u002Fcosmos_conf_1.png",{},"\u002Fpresentations\u002Fen\u002Fcosmos-db-conference",{"title":382,"description":498},"presentations\u002Fen\u002Fcosmos-db-conference","5min Lightning Talk","Cosmos DB Conference 2025","D5ytOTx5CBU","112","8gYT8h4-NH_SlCH08OUyQLQX1esO8TZcr0hIxWnIPmU",{"id":513,"title":514,"audience":515,"body":516,"carouselItems":136,"companyName":136,"date":136,"description":617,"end_date":136,"eventDate":618,"eventDuration":619,"eventName":620,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":621,"mainTitle":136,"meta":622,"navigation":144,"path":623,"podcastId":136,"role":136,"seo":624,"series":136,"start_date":136,"stats":136,"stem":625,"subtitle":626,"tags":136,"techs":136,"titleHighlight":627,"videoId":136,"views":136,"watchTime":136,"__hash__":628},"presentations\u002Fpresentations\u002Fen\u002Ffhnw_ai_agent_platforms.md","AI Agents in Production","ca. 35 participants",{"type":8,"value":517,"toc":607},[518,521,524,527,531,534,537,541,544,547,551,554,557,561,564,567,571,574,577,581,584,587,591,594,597,601,604],[11,519,520],{},"On 13 March 2026 I had the opportunity to give a talk at the FHNW AI Agents Bootcamp on the topic of \"AI Agents in Production\". The central question was: what must an AI platform deliver so that agents do not just work as a proof of concept but scale reliably in everyday business operations?",[11,522,523],{},"The talk opened with a concrete use case: key people leave the company and their knowledge goes with them. An AI assistant answers around 80 percent of queries directly from manuals and the knowledge base. Where that is not sufficient, an AI agent steps in, forwards the question to an expert and stores the new knowledge permanently for everyone. This setup works well for a single agent. But as soon as a second or third agent is added, the so-called day-two problem emerges: a triplicated knowledge base, triplicated access control, triplicated testing, triplicated model integration. That does not scale.",[11,525,526],{},"The road traffic analogy captures it well: a single car drives fine without traffic lights. But a hundred cars need traffic rules, signals, surveillance and a roadworthiness check, not faster engines. From the third use case onward, a platform is needed to solve these cross-cutting concerns once centrally rather than replicating them for every use case.",[19,528,530],{"id":529},"module-1-knowledge-base-connecting-company-knowledge","Module 1: Knowledge Base - Connecting Company Knowledge",[11,532,533],{},"The first building block is the central knowledge base. The problem is straightforward: a service agent must answer from manuals, tickets and process documents, but the language model does not know this content. Retrieval-Augmented Generation (RAG) closes that gap. Documents are split into small sections and stored as vectors. An incoming question is also vectorised, and the system finds semantically similar sections to pass as context to the model. The result: the model answers with company knowledge it never saw during training.",[11,535,536],{},"Today each agent often maintains its own database, documents are processed twice over and results are inconsistent. A central knowledge base with clearly defined areas solves this once for all. New agents are immediately connected with no additional effort.",[19,538,540],{"id":539},"module-2-governance-who-can-do-what","Module 2: Governance - Who Can Do What?",[11,542,543],{},"The second module addresses a question that is often asked too late in practice: under whose identity does an agent operate when it accesses customer data, process documents and technical manuals? There are two fundamentally different approaches. The agent as assistant inherits the permissions of the logged-in user, like an intern working under a colleague's login with the user's maximum access. The agent as an independent entity, by contrast, has its own identity and its own rights, precisely the minimal permissions it needs for its task.",[11,545,546],{},"Agents as independent entities make AI governable: they follow the least-privilege principle, can work without a user context (for example in overnight batch processing or automatic ticket responses), deliver consistent inputs for systematic testing and leave clear, attributable traces and logs.",[19,548,550],{"id":549},"module-3-transparency-what-happened","Module 3: Transparency - What Happened?",[11,552,553],{},"Imagine a service agent gives a technician the wrong repair instructions. The customer is angry. What happens next? Without transparency the answer is: we have no idea how this could have happened. No audit trail, no traceability, and trust is lost. With transparency you can reconstruct what happened: which documents were consulted, how the context was assembled, what the model produced and which intermediate steps were taken in longer workflows.",[11,555,556],{},"I posed a simple question to participants: would you run a business process that nobody can trace? Transparency is not a nice-to-have. It is the prerequisite for responsible AI operations.",[19,558,560],{"id":559},"module-4-evaluation-testing-ai-like-software","Module 4: Evaluation - Testing AI Like Software",[11,562,563],{},"Transparency logs are valuable not only for operations; they are the foundation for systematic quality assurance. AI quality is measurable, not a matter of luck. With a reference dataset of typical questions and verified answers, an LLM-as-judge approach (one language model evaluates the response of another) and custom evaluators for domain-specific criteria such as source correctness, compliance and tone, quality can be measured automatically. After every change, whether a new prompt, new documents or a new model, evaluation runs and immediately shows whether quality has risen or fallen.",[11,565,566],{},"One important point: agents with their own identity deliver consistent inputs and are what make systematic evaluation possible in the first place. The modules build on one another.",[19,568,570],{"id":569},"module-5-llm-gateway-model-changes-without-chaos","Module 5: LLM Gateway - Model Changes Without Chaos",[11,572,573],{},"Models are changed more often than you might think: because a provider deprecates a version, because a newer model offers better quality, because costs have dropped significantly or for legal reasons. Without a gateway, every agent is hard-wired to a specific model. As the number of agents grows, a necessary model switch quickly turns into chaos, especially when speed matters.",[11,575,576],{},"The LLM gateway decouples agents from specific models. Use cases address logical endpoints such as \"thinking-large\", \"fast-small\" or \"embedding\", and the gateway maps these to the currently active concrete models. A service agent, an onboarding agent and a compliance agent all communicate with the gateway, which routes requests to Claude Opus, GPT-4o mini or the embedding model depending on requirements. Cost control through central rate limits and budget caps is a further advantage of this architecture.",[19,578,580],{"id":579},"module-6-pii-protection-protecting-sensitive-data","Module 6: PII Protection - Protecting Sensitive Data",[11,582,583],{},"The sixth module addresses a central concern in the Swiss enterprise context: personal data must not leave the organisation uncontrolled. A service agent handling customer queries deals with names, email addresses, customer numbers and sometimes credit card data. Depending on the provider, requests may be stored or used for training purposes; GDPR and the Swiss Data Protection Act require the protection of personal data; and even without storage, the question remains whether sensitive data should be transmitted over the internet at all.",[11,585,586],{},"The solution: PII protection at gateway level, automatic, central and without any code changes in the individual agents. Before a request leaves the organisation, sensitive data is detected and handled: names are masked with placeholders, email addresses and IBANs are replaced, and credit card data blocks the request entirely. Solved once centrally, the same pattern applies as with all other modules.",[19,588,590],{"id":589},"how-the-modules-work-together","How the Modules Work Together",[11,592,593],{},"The six modules only realise their full value in combination. A complete request through the platform looks like this: the service technician asks a question. RAG searches the central knowledge base. Governance checks whether this agent is permitted to access the relevant data. Transparency logs which documents were consulted and which prompt was assembled. PII protection masks sensitive data in the request. The LLM gateway routes to the right model. And evaluation regularly checks whether quality is still on target.",[11,595,596],{},"Each module is built once and then works for all agents. That is the decisive principle: solve it once centrally, rather than replicating it for every use case.",[19,598,600],{"id":599},"synthesis-the-platform-decides-whether-ai-scales","Synthesis: The Platform Decides Whether AI Scales",[11,602,603],{},"The closing question of the talk brings everything together: do we solve this problem once centrally, or do we repeat it for every use case? A PoC is quick to build. But the platform determines whether AI actually scales in the enterprise.",[11,605,606],{},"The discussion prompts at the end sparked lively conversations: which module would you build first? How would you describe your current AI agents, as assistants or as independent entities? And what happens in your organisation today when an AI system makes a mistake? These questions struck a nerve with participants and showed that the path from successful proof of concepts to production-ready AI still lies ahead for many organisations.",{"title":124,"searchDepth":125,"depth":125,"links":608},[609,610,611,612,613,614,615,616],{"id":529,"depth":128,"text":530},{"id":539,"depth":128,"text":540},{"id":549,"depth":128,"text":550},{"id":559,"depth":128,"text":560},{"id":569,"depth":128,"text":570},{"id":579,"depth":128,"text":580},{"id":589,"depth":128,"text":590},{"id":599,"depth":128,"text":600},"On 13 March 2026 at the FHNW AI Agents Bootcamp I showed why a single working agent is not yet a solution\nand why a central platform becomes unavoidable from the third use case onward. Using six modules, knowledge\nbase, governance, transparency, evaluation, LLM gateway and PII protection, I explained how each of these\ncross-cutting concerns can be solved once centrally rather than replicated for every use case. The central\nmessage: a PoC is quick to build, but the platform determines whether AI actually scales in the enterprise.\n","2026-03-13","ca 40 Min.","Bootcamp AI-Agents - Transforming Organizations with AI Agents","\u002Fimages\u002Fpresentations\u002Ffhnw\u002Fai_agent_platform.png",{},"\u002Fpresentations\u002Fen\u002Ffhnw_ai_agent_platforms",{"title":514,"description":617},"presentations\u002Fen\u002Ffhnw_ai_agent_platforms","What the Platform Must Deliver","Transforming Organizations with AI Agents","nKa8LzbEk9n6JBC9cUFTeeAL-ZS4J88MEcu0vJGhz44",{"id":630,"title":631,"audience":632,"body":633,"carouselItems":136,"companyName":136,"date":136,"description":713,"end_date":136,"eventDate":714,"eventDuration":619,"eventName":715,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":716,"mainTitle":136,"meta":717,"navigation":144,"path":718,"podcastId":136,"role":136,"seo":719,"series":136,"start_date":136,"stats":136,"stem":720,"subtitle":721,"tags":136,"techs":136,"titleHighlight":722,"videoId":136,"views":136,"watchTime":136,"__hash__":723},"presentations\u002Fpresentations\u002Fen\u002Ffhnw_ai_trends.md","Overview of Current AI Trends in Swiss Industry","20 participants",{"type":8,"value":634,"toc":706},[635,638,641,645,648,651,655,658,661,664,667,671,674,677,680,684,687,690,693,697,700,703],[11,636,637],{},"On 5 March 2026 I had the opportunity to give a talk at the FHNW AI Bootcamp on \"Overview of Current AI Trends in Swiss Industry\". Under the motto \"Four Trends, One Pattern\" my aim was not simply to give participants a snapshot of the AI landscape, but to make visible the hidden logic behind seemingly independent developments. My promise to the audience was clear: you will leave the room with a clear picture of how these trends connect, and with three concrete questions for your next AI decision.",[11,639,640],{},"The talk opened with a number that surprises many: AI has become 300 times cheaper over the past 18 months. The price per AI intelligence unit fell from 32 to 0.10 US dollars. That is not incremental progress. It is a structural shift that makes everything that follows possible.",[19,642,644],{"id":643},"trend-1-model-revolution-the-choice-becomes-low-risk","Trend 1: Model Revolution - The Choice Becomes Low-Risk",[11,646,647],{},"The first trend concerns the models themselves. GPT, Claude, Gemini, DeepSeek: the leading frontier models now operate at comparable levels. This is shown by both the Artificial Analysis Intelligence Index and independent benchmarks such as the Mensa Norway IQ Test from trackingai.org. Proprietary models like GPT (OpenAI), Claude (Anthropic) and Gemini (Google) face capable open-source alternatives such as DeepSeek, Llama (Meta) and Qwen (Alibaba).",[11,649,650],{},"The practical consequence is decisive: companies no longer need to commit to a single model. All of them are good enough. Model selection has become low-risk. This creates a clear architecture: the model layer (data centre with proprietary and open-source models) becomes interchangeable infrastructure, comparable to the cloud ten years ago.",[19,652,654],{"id":653},"trend-2-agents-from-chat-to-action","Trend 2: Agents - From Chat to Action",[11,656,657],{},"The second trend is the shift from conversation partner to co-worker. A chatbot answers the question \"What were our Q4 numbers?\" An agent, by contrast, gets the task done: generate the quarterly report, compare it with the previous year, send a draft to management. Agents can use tools, follow workflows and ask clarifying questions. But an agent without access to a company's systems is like a new employee without a laptop.",[11,659,660],{},"An important conceptual distinction that I often miss in practice: a reactive assistant helps the user. An enterprise agent helps the organisation. It has its own identity, works for the organisation and receives task-based access to data, not user-dependent access. An agent that only responds to user requests is not an agent; it is an assistant.",[11,662,663],{},"On the spectrum between predefined workflows and fully autonomous systems, my practical recommendation is to start with workflows and extend incrementally. Predefined workflow agents are predictable and controllable, ideal for familiar, standardised processes. Agentic AI plans and decides autonomously, making it more powerful but harder to steer.",[11,665,666],{},"As a concrete practical example I showed a knowledge management use case: key people retire and their knowledge goes with them. An AI assistant covers around 80 percent of queries from existing manuals and documents. Where the knowledge base falls short, an AI agent forwards the question to an expert and stores the new answer permanently in the knowledge base. The result is a learning system that preserves institutional knowledge before it leaves the organisation.",[19,668,670],{"id":669},"trend-3-integration-connecting-ai-islands","Trend 3: Integration - Connecting AI Islands",[11,672,673],{},"The third trend is MCP, the Model Context Protocol. I like to describe it as \"USB for AI\": an open standard connecting AI platforms (as MCP clients) to any enterprise software (as MCP servers), including CRM, ERP, calendars, documents and ticketing systems. On one side are systems like Claude Desktop or Claude Code; on the other are PostgreSQL, GDrive, Git, Slack and Google Maps.",[11,675,676],{},"The second practical example in this section comes from an industrial context: a machine manufacturer has analytical sensor data, deterministic but context-free. Only the combination with usage and maintenance data from the customer, accessible via MCP, enables real value: predictive maintenance and auto-calibration. The value lay not in the model; it lay in the connection.",[11,678,679],{},"Another example: a device manufacturer automates support by having an AI agent work directly in Jira. When a ticket is opened, the agent automatically searches manuals and past tickets, creates a draft response and submits it to the support team for approval. The agent works where the team works, not in a separate AI tool. The complete picture shows a three-layer architecture: business applications (CRM, ERP, analytics) connected via API, MCP and A2A to the AI platform, which in turn accesses the data centre via API.",[19,681,683],{"id":682},"trend-4-lock-in-the-flip-side-of-the-platform","Trend 4: Lock-In - The Flip Side of the Platform",[11,685,686],{},"The fourth trend is the strategic flip side of platform development. What started as a simple chatbot in 2023 (ask questions, get answers) became an assistant in 2024 (analyse documents, understand images) and by 2025\u002F26 has become a full platform that books hotels, manages calendars and completes tasks autonomously. The more powerful the platform, the greater the dependency.",[11,688,689],{},"Lock-in creeps in at three levels. The knowledge base, with documents, vectorisation and synchronisation, is nearly impossible to migrate. Workflows built in visual builders are generally not exportable. And the memory, chat histories, preferences, context knowledge, is simply lost when switching platforms.",[11,691,692],{},"The good news: sovereignty is possible at every level. Open standards such as MCP and A2A protect against dependency. And for the Swiss context, a look at the Swiss AI Hub and providers such as Infomaniak shows that realistic alternatives to full dependence on US hyperscalers exist, from the model layer to the platform level.",[19,694,696],{"id":695},"synthesis-it-is-not-the-model-that-decides-it-is-the-platform","Synthesis: It Is Not the Model That Decides; It Is the Platform",[11,698,699],{},"The four trends follow one pattern: models converge and the choice becomes low-risk. Agents act, but they need tools and data. MCP connects as an open standard. And lock-in arises because the platform choice is strategic. The central message of the talk can be summarised in one sentence: it is not the model that decides. The platform decides.",[11,701,702],{},"What does that mean in practice? Three questions companies should ask before choosing an AI platform: is it open? Who controls the data? What integrations are possible, today and in the future?",[11,704,705],{},"The discussion afterwards showed that these questions struck a nerve with many participants, particularly the distinction between assistant and agent, and the concrete approach of starting with workflows rather than aiming for maximum autonomy immediately. I am convinced we are in a phase where the strategic decisions shaping AI use over the coming years are being made, and I look forward to continuing to be part of that journey.",{"title":124,"searchDepth":125,"depth":125,"links":707},[708,709,710,711,712],{"id":643,"depth":128,"text":644},{"id":653,"depth":128,"text":654},{"id":669,"depth":128,"text":670},{"id":682,"depth":128,"text":683},{"id":695,"depth":128,"text":696},"On 5 March 2026 at the FHNW AI Bootcamp I presented four current AI trends in Swiss industry: models are\nconverging and can now be chosen with low risk, agents are turning AI from a conversation partner into a\nco-worker, and open standards such as MCP connect AI to existing enterprise systems. Using concrete practical\nexamples from knowledge management and predictive maintenance I showed that the value lies not in the model\nbut in connecting it to your own data and systems. The central message: it is not the model that decides,\nit is the platform, because that is where long-term dependencies or genuine digital sovereignty arise.\n","2026-03-05","Bootcamp Artificial Intelligence for Management","\u002Fimages\u002Fpresentations\u002Ffhnw\u002Fai_trends.png",{},"\u002Fpresentations\u002Fen\u002Ffhnw_ai_trends",{"title":631,"description":713},"presentations\u002Fen\u002Ffhnw_ai_trends","Four Trends, One Pattern","Artificial Intelligence for Management","WaxwrUc16-hdpWYXb1pvKrknTYnTl9vhZA6UACZd60s",{"id":725,"title":726,"audience":727,"body":728,"carouselItems":814,"companyName":136,"date":136,"description":818,"end_date":136,"eventDate":819,"eventDuration":820,"eventName":821,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":822,"mainTitle":136,"meta":823,"navigation":144,"path":824,"podcastId":136,"role":136,"seo":825,"series":136,"start_date":136,"stats":136,"stem":826,"subtitle":827,"tags":136,"techs":136,"titleHighlight":828,"videoId":136,"views":136,"watchTime":136,"__hash__":829},"presentations\u002Fpresentations\u002Fen\u002Fhtcs.md","An Inspiring Start to the Future of the Medtech Sector","ca. 40 participants",{"type":8,"value":729,"toc":808},[730,733,737,740,767,770,772,776,779,782,786,789,791],[11,731,732],{},"At the recent New Year Aperitif of the Health Tech Cluster Switzerland I had the opportunity to give an engaging talk on the use of artificial intelligence in the healthtech sector. In a relaxed, almost intimate atmosphere, together with inspiring personalities and engaged experts, we kicked off the new year on an optimistic note.",[19,734,736],{"id":735},"practical-insights-and-innovative-ai-applications","Practical Insights and Innovative AI Applications",[11,738,739],{},"The afternoon was shaped by an intense exchange between specialists and interested attendees. My talk focused on how AI can create real value in various areas of the healthcare sector. I presented several scenarios showing how modern technologies can be applied in practice:",[339,741,742,749,755,761],{},[342,743,744,748],{},[745,746,747],"strong",{},"Support for medical devices"," - AI can assist with the operation and fault diagnosis of medical devices, allowing users to work faster and more safely.",[342,750,751,754],{},[745,752,753],{},"Documentation tools for nursing staff"," - Automated processes ease the time-consuming task of documentation, freeing up nursing staff for direct patient care.",[342,756,757,760],{},[745,758,759],{},"Information support for patients"," - Intelligent systems deliver tailored information that helps patients resolve their health questions more quickly.",[342,762,763,766],{},[745,764,765],{},"Compliance support for medical device manufacturers"," - Given constantly changing standards and regulations, AI solutions can help design compliant processes and meet legal requirements efficiently.",[11,768,769],{},"The goal of the event was to bridge theory and practice and show how AI creates value across different business areas.",[195,771],{":height":197,":items":198,":width":319},[19,773,775],{"id":774},"live-demos-and-direct-exchange","Live Demos and Direct Exchange",[11,777,778],{},"Another highlight of the event was the hands-on live demos at our test stations. Participants had the chance to experience and try out the AI applications showcased up close. One thing quickly became clear: innovation works best through direct dialogue and mutual learning.",[11,780,781],{},"Martin Egloff rounded off the afternoon with valuable insights into our company and our way of working, which further enriched the event. Personal networking and open exchange contributed significantly to making the aperitif not only informative but also inspiring.",[19,783,785],{"id":784},"outlook","Outlook",[11,787,788],{},"The evening showed once again: with hands-on approaches and innovative AI solutions, the future of the medtech sector can be actively shaped. I was pleased to map out this path and bring new impulses to the industry.",[19,790,337],{"id":336},[339,792,793,801],{},[342,794,795,800],{},[345,796,799],{"href":797,"rel":798},"https:\u002F\u002Fwww.linkedin.com\u002Fposts\u002Fhealth-tech-cluster-switzerland_healthtech-activity-7285989756029521920-z2DR\u002F?utm_source=share&utm_medium=member_android",[349],"Review of our New Year Aperitif (HTCS)",", LinkedIn.",[342,802,803,800],{},[345,804,807],{"href":805,"rel":806},"https:\u002F\u002Fwww.linkedin.com\u002Fposts\u002Fbbv-software-services_inspirationausderpraxis-netzwerken-medtech-activity-7286030193339625475-51wL\u002F?utm_source=share&utm_medium=member_android",[349],"Inspiration from practice and intimate networking (bbv)",{"title":124,"searchDepth":125,"depth":125,"links":809},[810,811,812,813],{"id":735,"depth":128,"text":736},{"id":774,"depth":128,"text":775},{"id":784,"depth":128,"text":785},{"id":336,"depth":128,"text":337},[815,816,817],"\u002Fimages\u002Fpresentations\u002Fhtcs\u002F1.jpg","\u002Fimages\u002Fpresentations\u002Fhtcs\u002F2.jpg","\u002Fimages\u002Fpresentations\u002Fhtcs\u002F3.jpg","At the New Year Aperitif of the Health Tech Cluster Switzerland I presented practical AI use cases in the\nmedtech sector, from support for medical devices to documentation tools for nursing staff and compliance\ntools for medical device manufacturers. Live demos gave participants the chance to try out the innovative\nAI solutions directly and exchange ideas in a relaxed atmosphere. The inspiring evening laid the groundwork\nfor future developments and underscored how valuable hands-on insights and personal networking are for the\nsector.\n","2025-01-16","ca. 20 Min.","Health Tech Cluster Switzerland - New Year Aperitif","\u002Fimages\u002Fpresentations\u002Fhtcs\u002Fhtcs_3.png",{},"\u002Fpresentations\u002Fen\u002Fhtcs",{"title":726,"description":818},"presentations\u002Fen\u002Fhtcs","AI Use Cases in Medicine: From Diagnosis to Therapy","New Year Aperitif","1yeAmBUR6K0Ym6w7g4HjR1ef9xSHFPwDRcHg1kBKjWU",{"id":831,"title":832,"audience":833,"body":834,"carouselItems":136,"companyName":136,"date":136,"description":918,"end_date":136,"eventDate":919,"eventDuration":920,"eventName":921,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":922,"mainTitle":136,"meta":923,"navigation":144,"path":924,"podcastId":136,"role":136,"seo":925,"series":136,"start_date":136,"stats":136,"stem":926,"subtitle":148,"tags":136,"techs":136,"titleHighlight":927,"videoId":136,"views":136,"watchTime":136,"__hash__":928},"presentations\u002Fpresentations\u002Fen\u002Findustrieforum_2025.md","AI Challenge","250-300 participants",{"type":8,"value":835,"toc":909},[836,839,842,845,848,852,855,858,862,865,869,872,876,879,882,886,889,893,896,899,903,906],[11,837,838],{},"I started my pitch by holding up a notebook and saying: \"Imagine all my knowledge and experience were captured here. But unfortunately something like that does not really exist.\" So much implicit knowledge is lost in organisations when employees leave. That was exactly the problem I wanted to address.",[11,840,841],{},"My pitch was part of the 11th \"Industrieforum 2025\", held under the motto \"Technology, People, Culture in Harmony\". The forum provided a platform for inspiring exchange between industry leaders and over 25 exhibitors presenting the latest technologies and innovations. The goal of the event was to discuss digital transformation in harmony with human values and explore new ways of working together.",[11,843,844],{},"After my one-minute pitch, the audience voted on which talks should be given in the afternoon. I narrowly missed enough votes, so my presentation was not delivered live and is only available as a video. Even so, the pitch was a lot of fun and vividly showed how artificial intelligence can help preserve implicit knowledge and make it actively usable. This idea is central to the future of knowledge management.",[15,846],{":video-id":847},"s6qEXDkxvOg",[19,849,851],{"id":850},"securing-company-knowledge-with-artificial-intelligence-a-practical-example","Securing Company Knowledge with Artificial Intelligence - A Practical Example",[11,853,854],{},"In many companies, the most valuable knowledge is not in manuals or databases but in the minds of long-serving employees. This \"implicit expertise\" is quickly lost, however, once key people leave the company or move to other departments. The result: other employees suddenly lack the required knowledge, have to rethink the same problems from scratch and must work through time-consuming processes to find solutions. In the long run this means not only a loss of innovative capacity and efficiency, but can also cause considerable costs.",[11,856,857],{},"At the 11th Industrieforum 2025 on 4 June 2024 I gave a presentation together with bbv Software Services AG that focused precisely on this problem. Under the title \"Securing Company Knowledge with Artificial Intelligence\" we showed how the application of modern AI systems, in particular Large Language Models (LLMs), can help to capture, structure and make implicit knowledge accessible to all employees. The central approach is based on the idea of \"AI agents\", virtual assistants that interact in principle like colleagues: they answer questions, conduct research, consult experts where necessary and summarise newly gained information in a central location.",[19,859,861],{"id":860},"the-underlying-problem-implicit-knowledge-gets-lost","The Underlying Problem: Implicit Knowledge Gets Lost",[11,863,864],{},"Implicit knowledge is hard to grasp because it is not written down but exists \"between the lines\". An employee who has looked after a particular product or a specific process for years often knows exactly where the pitfalls lie and what measures must be taken to ensure quality. This knowledge is rarely documented. You simply know it from daily practice. When that key person leaves the company, however, this know-how is difficult to reconstruct. Moreover, remaining colleagues are often not familiar with the details of the processes and must reacquire the knowledge through laborious effort. The result: rising costs, longer production and development times and recurring errors.",[19,866,868],{"id":867},"the-ai-approach-knowledge-management-with-ai-agents","The AI Approach: Knowledge Management with AI Agents",[11,870,871],{},"An important part of the presentation was the demonstration of a use case from food production. The example revolved around the question of how to maintain the freshness of products throughout the entire delivery process. An AI agent can act here as a kind of \"virtual employee\". It receives the query: \"How do we ensure that our products stay fresh from production to delivery to the customer?\" If it turns out that the answer is not in its existing data, it forwards the question to a recognised expert. Once their answer is available, the AI agent stores this information in the knowledge database and also creates a glossary entry, in this example for the technical term \"Frost Flow\". This approach makes it possible not only to pass the knowledge acquired to those immediately involved, but to make it permanently available to all employees.",[19,873,875],{"id":874},"llms-and-prompting-how-artificial-intelligence-gains-language-understanding","LLMs and Prompting: How Artificial Intelligence Gains Language Understanding",[11,877,878],{},"The technical foundation for AI agents is provided by Large Language Models, trained on vast amounts of data. At their core, these models learn to predict the next word in a sentence. In doing so they develop a remarkable understanding of grammar, sentence structure and even complex relationships. In practice, however, they are only as good as the instructions they are given, the so-called \"prompting\". Here, for example, one must define from which perspective the model should respond, what information is available to it and what kind of answer is desired (particularly detailed, or short and concise).",[11,880,881],{},"In addition to this basic structure, \"Retrieval Augmented Generation (RAG)\" can be used to integrate company-specific databases. This means the agent draws not only on its internal training but actively accesses current and specific knowledge from within the organisation. This interplay makes it possible for answers to be not only linguistically correct but also to incorporate the concrete context of the company and deliver relevant details.",[19,883,885],{"id":884},"multi-modal-interaction-and-tailored-answers","Multi-Modal Interaction and Tailored Answers",[11,887,888],{},"Furthermore, modern AI systems are capable of handling not only text data but also speech, images or other files. For example, the agent could evaluate an uploaded photo or analyse a voice message to derive context-based answers. Also interesting is the possibility of presenting the same information adapted to different roles or departments. While an operations manager might need a precise technical description, an environmental officer may only require a brief summary focused on environmental aspects.",[19,890,892],{"id":891},"costs-effort-and-opportunities","Costs, Effort and Opportunities",[11,894,895],{},"Of course, introducing such AI systems involves considerable effort. Companies must first decide whether to train their own model or use an existing model and fine-tune it for their needs. Both options require investment, whether in hardware, AI expertise or high-quality data intended to optimise the model. Nevertheless, early experience shows that these efforts pay off in the long run, particularly when knowledge is thereby secured permanently and duplicate work is reduced.",[11,897,898],{},"Especially in times when workforces change rapidly, securing company knowledge is crucial for maintaining competitiveness and a drive for innovation. AI agents that collect, structure and deliver knowledge on demand can make a decisive contribution to this.",[19,900,902],{"id":901},"conclusion","Conclusion",[11,904,905],{},"With my talk at the 11th Industrieforum 2025, together with bbv Software Services AG, I was able to show that AI agents are not merely a theoretical concept but deliver tangible results in practice. From capturing implicit knowledge through creating glossary entries to delivering context-sensitive answers: modern AI systems offer companies a powerful tool for preserving know-how and making it quickly accessible to new employees. This prevents valuable experiential knowledge from being lost the moment an employee walks out the door.",[11,907,908],{},"Artificial intelligence thus offers not only potential for process automation and data analysis, but also opens up an entirely new perspective on knowledge management, a perspective that places people at the centre and ensures that their knowledge and experience are not lost but instead become a central resource from which the entire organisation benefits.",{"title":124,"searchDepth":125,"depth":125,"links":910},[911,912,913,914,915,916,917],{"id":850,"depth":128,"text":851},{"id":860,"depth":128,"text":861},{"id":867,"depth":128,"text":868},{"id":874,"depth":128,"text":875},{"id":884,"depth":128,"text":885},{"id":891,"depth":128,"text":892},{"id":901,"depth":128,"text":902},"At the 11th Industrieforum 2025 I presented my vision for AI-supported knowledge management that preserves\nimplicit knowledge and makes it efficiently usable through context-based search methods such as Retrieval\nAugmented Generation (RAG). I described AI agents as virtual employees who not only structure knowledge but\ncan also generate new insights in real time. Despite challenges such as information overload and data\nprotection, I showed how AI offers transformative possibilities for a future-ready knowledge culture.\n","2024-06-04","1 Min (18 Min.)","11. «Industrieforum 2025»","\u002Fimages\u002Fpresentations\u002Findustrieforum_2025\u002Findustrieforum.png",{},"\u002Fpresentations\u002Fen\u002Findustrieforum_2025",{"title":832,"description":918},"presentations\u002Fen\u002Findustrieforum_2025","«Industrieforum 2025»","D7eQJv2xvS3uOK9ErWfW64PDEHHhquSShTxeZKNlXqw",{"id":930,"title":931,"audience":932,"body":933,"carouselItems":1041,"companyName":136,"date":136,"description":1044,"end_date":136,"eventDate":1045,"eventDuration":820,"eventName":1046,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":1047,"mainTitle":136,"meta":1048,"navigation":144,"path":1049,"podcastId":136,"role":136,"seo":1050,"series":136,"start_date":136,"stats":136,"stem":1051,"subtitle":1052,"tags":136,"techs":136,"titleHighlight":1053,"videoId":136,"views":136,"watchTime":136,"__hash__":1054},"presentations\u002Fpresentations\u002Fen\u002Fnlp-expert-group.md","AI Agent Collaboration","> 60 participants",{"type":8,"value":934,"toc":1031},[935,939,942,947,951,954,959,963,966,971,975,978,983,987,990,995,999,1002,1007,1011,1014,1019,1023,1026],[19,936,938],{"id":937},"from-assistant-to-agent-when-ai-arrives-in-the-process","From Assistant to Agent: When AI Arrives in the Process",[11,940,941],{},"The term \"AI agent\" is used quickly, yet it often remains unclear whether everyone means the same thing. In my webinar I therefore start with a precise definition. A reactive assistant answers requests and delivers content \"alongside the process\". A process-integrated agent is designed to work within the workflow: it responds to events, pursues goals across multiple steps, uses tools and interacts with systems. What matters is not maximum autonomy but reliable execution within clear guardrails, so that results remain traceable.",[27,943,944],{},[11,945,946],{},"What are AI Agents? I guess we know, but do we agree?\nReactive Assistants vs. Process Integrated Agents",[19,948,950],{"id":949},"a-spectrum-rather-than-black-and-white-control-vs-freedom","A Spectrum Rather Than Black and White: Control vs. Freedom",[11,952,953],{},"Agents are rarely an either\u002For. I describe them as a spectrum between predefined workflows and fully autonomous systems. In practice the productive range often lies in between: depending on risk and complexity, \"more freedom\" is not automatically better. An agent is useful when it has enough room to get work done but enough control to remain stable, technically and organisationally.",[27,955,956],{},[11,957,958],{},"The Spectrum of Agents — Control vs. Freedom",[19,960,962],{"id":961},"modularity-as-a-principle-building-blocks-rather-than-a-monolith","Modularity as a Principle: Building Blocks Rather Than a Monolith",[11,964,965],{},"Another focus is modularity. When agents are built as monolithic all-rounders, they quickly become difficult to test and difficult to steer. I therefore think in \"building blocks\": clearly bounded components that can be combined without every extension destabilising the overall system. This is both an architecture principle and a team principle: capabilities remain interchangeable, responsibilities stay clear and further development becomes more plannable.",[27,967,968],{},[11,969,970],{},"Building Blocks — The Power of Modular Agents",[19,972,974],{"id":973},"role-model-taco-how-agents-act-in-collaboration","Role Model TACO: How Agents Act in Collaboration",[11,976,977],{},"To make functions in agentic systems tangible I use a role model (TACO). It helps distinguish agents by their function in the value chain: from focused \"Askers\" for single goals, through \"Automators\" for recurring tasks, to \"Collaborators\" for human-agent teamwork and \"Orchestrators\" that coordinate complex end-to-end processes across multiple systems. As orchestration grows, integration and operational requirements typically rise as well, making it increasingly necessary to design communication and control carefully.",[27,979,980],{},[11,981,982],{},"Focus on single goals, broken into simple steps. Are the easiest to manage.\nAre for human-agent teamwork, acting like teammates, not just tools\nManage complex, end-to-end processes across systems",[19,984,986],{"id":985},"multiple-agents-one-core-problem-communication","Multiple Agents, One Core Problem: Communication",[11,988,989],{},"As soon as multiple agents work together, communication becomes the central system question. I draw on the \"Agent Communication Trilemma\": efficiency, portability and versatility are in tension with one another. An approach that is very efficient for a specific environment often loses portability. A highly portable approach is frequently less optimised. And maximum versatility can increase the cost of standardisation and operations. The practical consequence follows: agents need an explicit language for tasks, states and results, appropriate to the system landscape and the operating model.",[27,991,992],{},[11,993,994],{},"The Agent Communication Trilemma\nEFFICIENCY — PORTABILITY — VERSATILITY",[19,996,998],{"id":997},"protocols-as-bridges-mcp-agent-protocol-and-chat-like-approaches","Protocols as Bridges: MCP, Agent Protocol and Chat-Like Approaches",[11,1000,1001],{},"I map several protocol directions that address precisely this gap between the model and the \"outside world\". The Model Context Protocol (MCP) stands out as an approach to standardising tool and context integration. Agentprotocol.ai describes work more strongly via tasks, steps and artefacts, in clear, process-oriented units. And chat-inspired protocols like the LangChain Agent Protocol lean more toward conversation and runtime concepts that many agent systems already use.",[27,1003,1004],{},[11,1005,1006],{},"Model Context Protocol (MCP) — Anthropic's approach\nJust Tasks, Steps and Artifacts?\nLangchain Agent Protocol — A chat-inspired approach",[19,1008,1010],{"id":1009},"agora-protocols-on-demand-natural-language-only-when-necessary","AGORA: Protocols on Demand, Natural Language Only When Necessary",[11,1012,1013],{},"As a conceptual addition I introduce AGORA: an approach that allows different protocol forms and uses appropriate structures depending on the situation, up to and including \"Protocol Documents\" that can be referenced. The guiding idea is pragmatic: natural language is flexible but hard to validate and version. It is therefore treated as a last resort rather than the default transport format for precise states.",[27,1015,1016],{},[11,1017,1018],{},"Natural language as a last resort\nAgents negotiate which protocol to use",[19,1020,1022],{"id":1021},"a-target-architecture-of-our-own-closed-event-based-observable","A Target Architecture of Our Own: Closed, Event-Based, Observable",[11,1024,1025],{},"In closing I condense the content into a practice-oriented target picture: a closed, event-based system that enables proactivity while consistently keeping humans in the loop. Transparency is a basic requirement: actions and events must be observable so that debugging, auditability and operational security become possible. Workflows are predominantly predefined, with AI acting mainly on routing and flexible execution within the guardrails; thread and run concepts serve as a structural orientation for the runtime.",[27,1027,1028],{},[11,1029,1030],{},"Transparency — Every action and event can be observed\nPredominantly predefined workflows with AI routing\nEvent-based system allowing for proactivity — Human in the Loop",{"title":124,"searchDepth":125,"depth":125,"links":1032},[1033,1034,1035,1036,1037,1038,1039,1040],{"id":937,"depth":128,"text":938},{"id":949,"depth":128,"text":950},{"id":961,"depth":128,"text":962},{"id":973,"depth":128,"text":974},{"id":985,"depth":128,"text":986},{"id":997,"depth":128,"text":998},{"id":1009,"depth":128,"text":1010},{"id":1021,"depth":128,"text":1022},[1042,1043],"\u002Fimages\u002Fpresentations\u002Fnlp_expert_meeting\u002F1.jpeg","\u002Fimages\u002Fpresentations\u002Fnlp_expert_meeting\u002F2.jpg","An AI agent is positioned as an evolution beyond reactive assistants and examined along a spectrum between\npredefined workflows and high autonomy. The focus is on how agents can be reliably integrated into processes\nthrough clear guardrails, modularity and observable execution. A role model (TACO) classifies agents by\nfunction, from focused single tasks through to orchestrating complex end-to-end workflows. Multi-agent\nsystems bring communication to the fore as the central challenge, including the tension between efficiency,\nportability and versatility. Various protocol approaches such as MCP, task- and artefact-oriented models\nand chat-like runtime concepts are compared as building blocks for structured collaboration between agents\nand systems.\n","2025-02-04","NLP Expert Group Meeting - AI Agents","\u002Fimages\u002Fpresentations\u002Fnlp_expert_meeting\u002Fnlp_experts.png",{},"\u002Fpresentations\u002Fen\u002Fnlp-expert-group",{"title":931,"description":1044},"presentations\u002Fen\u002Fnlp-expert-group","A roadmap for successful adoption and scaling in organizations","AI Agents","JI1CVotZZ3N92cLoQrpjrdCpEO-5GP0rlIJLWjZtX9E",{"id":1056,"title":149,"audience":1057,"body":1058,"carouselItems":1193,"companyName":136,"date":136,"description":1201,"end_date":136,"eventDate":1202,"eventDuration":820,"eventName":1203,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":1204,"mainTitle":136,"meta":1205,"navigation":144,"path":1206,"podcastId":136,"role":136,"seo":1207,"series":136,"start_date":136,"stats":136,"stem":1208,"subtitle":148,"tags":136,"techs":136,"titleHighlight":1209,"videoId":136,"views":136,"watchTime":136,"__hash__":1210},"presentations\u002Fpresentations\u002Fen\u002Fswico.md","50-80 participants",{"type":8,"value":1059,"toc":1185},[1060,1063,1066,1070,1073,1093,1095,1097,1101,1104,1109,1112,1115,1119,1122,1126,1129,1133,1136,1139,1143,1146,1149,1152,1156,1159,1162,1165,1167],[11,1061,1062],{},"On 19 April 2024 I, Marius Högger (AI Engineer at bbv), had the opportunity to give a talk on \"Knowledge Management with AI Agents\" at the industry association Swico's \"AI in Action\" event series. The event took place in a remarkable setting at Luma Westbau in Zurich, where floor-to-ceiling bookshelves and art installations created an inspiring environment for the exchange of knowledge.",[11,1064,1065],{},"Right from the start it was clear: Swico wanted not only to give participants theoretical background knowledge but to show them through concrete examples and practical insights how AI can be used effectively. Together with two other speakers, I was at the centre of a lively exchange that illuminated both the fascination and the challenges of artificial intelligence.",[19,1067,1069],{"id":1068},"the-swico-ai-event-series-ai-in-action","The SWICO AI Event Series \"AI in Action\"",[11,1071,1072],{},"The SWICO AI event series is designed to familiarise members and interested parties with current trends and developments in the field of artificial intelligence. Under the motto \"AI in Action\", participants on 19 April 2024 were able to follow engaging talks covering the following topics:",[339,1074,1075,1081,1087],{},[342,1076,1077,1080],{},[745,1078,1079],{},"\"ChatGPT Demystified\""," - presented by Joel Barmettler (Senior AI Engineer at bbv and Universität Zürich), who used machine learning fundamentals and concrete examples to show how Large Language Models (LLMs) like ChatGPT work and where their limits lie.",[342,1082,1083,1086],{},[745,1084,1085],{},"\"Knowledge Management with AI Agents\""," - my talk, in which I presented a proof of concept for the practical use of AI agents in an organisation. The focus was on how employees gain access to relevant expert knowledge through AI-supported systems without getting lost in floods of data.",[342,1088,1089,1092],{},[745,1090,1091],{},"\"Strategies for Transforming Ideas into AI-Powered Products\""," - delivered by my colleague Dr. Emre Özyurt (AI Consultant at bbv), who presented methods and approaches for turning early ideas into viable AI products.",[11,1094,769],{},[195,1096],{":height":197,":items":198,":width":319},[19,1098,1100],{"id":1099},"my-talk-knowledge-management-with-ai-agents","My Talk: \"Knowledge Management with AI Agents\"",[11,1102,1103],{},"In my presentation I asked how we can better navigate the ubiquitous \"knowledge labyrinth\" in organisations with the help of AI agents. Important information is often scattered across different departments, databases and documents, which means a time-consuming search, especially for new employees.",[1105,1106,1108],"h4",{"id":1107},"context-is-the-key","Context Is the Key",[11,1110,1111],{},"A core element of my presentation was the importance of context. Classic chatbots like ChatGPT are based on statistical language models and excel at processing natural language. However, their factual accuracy is not always guaranteed, since they frequently only predict the most probable next word without necessarily drawing on real data from the company environment. To address this problem, I relied on the principle of Retrieval Augmented Generation (RAG).",[11,1113,1114],{},"With RAG, external company-specific data sources are dynamically incorporated into the response process. This means the AI, rather than relying solely on its training, retrieves relevant documents from a knowledge database and extracts concrete information from them. This step substantially improves the quality of answers and ensures that the AI system understands the company context and delivers fact-based results when needed.",[1105,1116,1118],{"id":1117},"ai-agents-as-virtual-employees","AI Agents as Virtual Employees",[11,1120,1121],{},"A further focus was on the concept of AI agents. Unlike pure chatbots, agents have a defined \"role\" and \"task\"; they act like virtual employees who not only answer questions but actively absorb expert knowledge from real employees and make it available again in real time. I explained how such agents can be configured using role profiles and \"character traits\" (for example tone, area of expertise, responsibility) so that they respond precisely to the problem at hand.",[1105,1123,1125],{"id":1124},"proof-of-concept-value-in-practice","Proof of Concept: Value in Practice",[11,1127,1128],{},"To make my explanations tangible, I presented a proof of concept in which an AI agent supports newly hired employees during onboarding. Instead of laboriously working through countless documents, they receive direct access to relevant process documents, checklists and contact persons via a simple chat interface. The agent learns continuously from employee inputs, stores new insights and makes them automatically available the next time they are needed.",[19,1130,1132],{"id":1131},"the-fire-alarm-an-unplanned-break","The Fire Alarm: An Unplanned Break",[11,1134,1135],{},"During my talk something happened that literally \"heated up\" the evening: while preparing the aperitif in the adjacent room, something apparently caught fire, triggering an unplanned fire alarm. Suddenly loud sirens sounded and we had to vacate the event hall in a hurry.",[11,1137,1138],{},"What initially caused a little disruption turned out in hindsight to be a welcome opportunity to let the material sink in. Some participants used the brief enforced pause to discuss possible AI use cases in their own companies outside. After a few minutes the all-clear was given and we were able to continue the presentation and the rest of the event without further incident.",[19,1140,1142],{"id":1141},"discussions-and-outlook","Discussions and Outlook",[11,1144,1145],{},"In the subsequent discussion round many of those present still had open questions on data protection, permissions and bias. Topics included how much access an AI agent should have to confidential company information and how to ensure that no unwanted data leaks occur. I pointed out the possibility of controlling access rights at a granular level and only \"feeding\" agents with the information they actually need for their task.",[11,1147,1148],{},"The risk of manipulation and bias was also an important topic: since language models depend heavily on the available data, one-sided or incomplete datasets can easily lead to distortions (biases) in the AI's responses. This is precisely where measures such as controlled data management, careful prompt engineering and continuous monitoring and updating of models come in.",[11,1150,1151],{},"Beyond the technical challenges, there is always the question of how to finance such a project, justify it to management and integrate it into existing organisational structures in the long term. Here my colleague Dr. Emre Özyurt highlighted in his talk that it is often advisable to start with a pilot project in a limited knowledge area. Once the results are convincing, further knowledge domains can be opened up step by step.",[19,1153,1155],{"id":1154},"conclusion-ai-agents-in-the-enterprise","Conclusion: AI Agents in the Enterprise",[11,1157,1158],{},"The response to my presentation at the Swico AI event series was consistently positive. Participants saw great potential in using AI agents in their own companies to optimise knowledge management and relieve employees of recurring research tasks. Despite occasional concerns around data protection and costs, many agreed that a well-conceived and tightly monitored AI system delivers enormous advantages in the long run.",[11,1160,1161],{},"Knowledge agents have the potential to revolutionise access to company knowledge: they act proactively, learn continuously and create a dynamic, living knowledge network. For companies operating in a highly competitive environment, this can be a decisive competitive advantage.",[11,1163,1164],{},"Overall the event, unplanned fire alarm included, was a complete success: it conveyed a solid understanding of current AI technologies and a clear picture of how they can be transferred effectively into business practice. For me personally it was an exciting opportunity to present my proof of concept to an interested specialist audience and receive valuable feedback. I am convinced that we will see further rapid advances in the coming years and look forward to continuing this journey together with the Swico community and interested organisations.",[19,1166,337],{"id":336},[339,1168,1169,1177],{},[342,1170,1171,1176],{},[345,1172,1175],{"href":1173,"rel":1174},"https:\u002F\u002Fwww.swico.ch\u002Fde\u002Fnews\u002Fdetail\u002Fki-in-aktion-transformation-von-wissensmanagement-durch-kunstliche-intelligenz",[349],"AI in Action: Transformation of Knowledge Management through Artificial Intelligence",", on SWICO.ch.",[342,1178,1179,1184],{},[345,1180,1183],{"href":1181,"rel":1182},"https:\u002F\u002Fwww.netzwoche.ch\u002Fnews\u002F2024-04-22\u002Frauchende-koepfe-und-kuechen-beim-swico",[349],"Smoking heads and kitchens at Swico"," from Netzwoche.",{"title":124,"searchDepth":125,"depth":125,"links":1186},[1187,1188,1189,1190,1191,1192],{"id":1068,"depth":128,"text":1069},{"id":1099,"depth":128,"text":1100},{"id":1131,"depth":128,"text":1132},{"id":1141,"depth":128,"text":1142},{"id":1154,"depth":128,"text":1155},{"id":336,"depth":128,"text":337},[1194,1195,1196,1197,1198,1199,1200],"\u002Fimages\u002Fpresentations\u002Fswico\u002Fimg_2.png","\u002Fimages\u002Fpresentations\u002Fswico\u002Fimg_3.png","\u002Fimages\u002Fpresentations\u002Fswico\u002Fimg_4.png","\u002Fimages\u002Fpresentations\u002Fswico\u002Fimg_5.png","\u002Fimages\u002Fpresentations\u002Fswico\u002Fimg_6.png","\u002Fimages\u002Fpresentations\u002Fswico\u002Fimg_7.png","\u002Fimages\u002Fpresentations\u002Fswico\u002Fimg_8.png","Knowledge management with AI agents makes it possible to navigate the \"knowledge labyrinth\" in organisations\nby connecting generative language models with company-specific data sources. Virtual employees (agents) can\nsupport new staff in finding information and make expert knowledge accessible in real time. At a Swico event\nthis solution was presented using a proof of concept, interrupted by a fire alarm that gave everyone a short\nbreather.\n","2024-04-19","SWICO - AI in Action","\u002Fimages\u002Fpresentations\u002Fswico\u002Fwissensmanagement.png",{},"\u002Fpresentations\u002Fen\u002Fswico",{"title":149,"description":1201},"presentations\u002Fen\u002Fswico","AI in Action","ETgrhwh3G5it4mwqo5Dbr75s2gzuVSWd2XMZLsd9CYg",{"id":1212,"title":1213,"audience":1214,"body":1215,"carouselItems":136,"companyName":136,"date":136,"description":1322,"end_date":136,"eventDate":1323,"eventDuration":1324,"eventName":1325,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":1326,"mainTitle":136,"meta":1327,"navigation":144,"path":1328,"podcastId":136,"role":136,"seo":1329,"series":136,"start_date":136,"stats":136,"stem":1330,"subtitle":1331,"tags":136,"techs":136,"titleHighlight":1332,"videoId":1333,"views":1334,"watchTime":136,"__hash__":1335},"presentations\u002Fpresentations\u002Fen\u002Fwebinar_agent_rag.md","Webinar: Agent-RAG","ca. 20 participants",{"type":8,"value":1216,"toc":1314},[1217,1220,1222,1225,1230,1234,1237,1240,1244,1247,1250,1255,1258,1261,1265,1268,1271,1275,1278,1283,1286,1290,1293,1296,1300,1303,1306,1311],[11,1218,1219],{},"In my most recent webinar I spoke together with my colleague Martin König and our moderator Stefan Heberling about how Retrieval Augmented Generation (RAG) works in practice. We wanted to show what technical details lie behind this approach and why it is so valuable for so many applications. The main focus was on the interplay between the various pipelines that allow an AI model to draw on company-specific or publicly available data and use it intelligently.",[15,1221],{":video-id":17},[11,1223,1224],{},"Martin opened the webinar by noting that we would be using more technical terminology than usual and would therefore spend time on basic explanations at the start. This covered above all the Large Language Model (LLM), which is at the heart of every AI agent, and the concept of the system prompt, which defines how the model's responses are formulated. We also made clear that the query, the user's question, and the additional context from a knowledge database must work together to deliver a comprehensive answer.",[27,1226,1227],{},[11,1228,1229],{},"\"A Large Language Model is the core of an AI agent: it generates continuous text by meaningfully extending what has already been said, forming the basis for every generative application.\"",[19,1231,1233],{"id":1232},"overview-of-the-pipelines","Overview of the Pipelines",[11,1235,1236],{},"The core of RAG can be broadly divided into three sections. These so-called pipelines determine how information is ingested, prepared and ultimately used. The first pipeline, the Ingestion Pipeline, is concerned with selecting documents from various sources, cleaning them and structuring them. These data are then stored in a knowledge database that can make them quickly available again later.",[11,1238,1239],{},"The second pipeline, the Query Pipeline, handles everything that happens when a user asks a question. It examines what the user actually wants and how their query might be reformulated to yield more relevant results. The third pipeline, the Retrieval Pipeline, then searches the knowledge database for the appropriate answers by comparing the vector of the user's question with the vectors of the chunks. The most similar chunks are initially selected because they match best in terms of content, and together with the original question they are processed into a fluent, comprehensible text.",[19,1241,1243],{"id":1242},"the-ingestion-pipeline","The Ingestion Pipeline",[11,1245,1246],{},"The first step in a RAG system involves reading and processing data before it goes into a knowledge database. During this step a decision is made about which documents to include at all and whether access permissions need to be taken into account. If internal policies prohibit certain files from being shared with an AI system, that content can be filtered out early on.",[11,1248,1249],{},"Once the right data have been selected, they are split into small sections. This process is usually referred to as chunking and its purpose is to divide knowledge in such a way that the model can later retrieve exactly the right excerpts. A simple approach is to split into fixed lengths, say every few thousand words. Often, however, it makes more sense to preserve the structure of a text and cut at headings or thematic sections. This ensures that a section remains coherent and does not mix with entirely different topics.",[27,1251,1252],{},[11,1253,1254],{},"\"A perfectly crafted text splitter divides the document the way we as humans would: at headings, at coherent paragraphs or at clearly delineated text boxes.\"",[11,1256,1257],{},"In the webinar we also pointed out that standard tools often do not know which elements of a document are superfluous. They sometimes include navigation menus, links and captions in awkward places, producing fragments of limited usefulness. A tailored solution that recognises exactly which content to use and splits it sensibly into chunks can remedy this.",[11,1259,1260],{},"Once the sections are created and cleaned, each one can be converted into a numerical vector using a language model. These vectors later reside in the knowledge database and enable semantic search. Metadata such as categories, access permissions or timestamps can supplement the content. This simplifies handling different document versions later and prevents confidential information from reaching unauthorised users.",[19,1262,1264],{"id":1263},"the-query-pipeline","The Query Pipeline",[11,1266,1267],{},"As soon as a user asks a question, the Query Pipeline springs into action. An important topic here is intent recognition. We want to understand what lies behind the question. Has a document been uploaded and simply needs to be summarised? Or does someone want to search sources stored in the company network? One simple approach is to feed a language model with example sentences and let it decide which category the request falls into. More complex scenarios can be handled using a dedicated classification model.",[11,1269,1270],{},"In some cases it also makes sense to rewrite the query, the user's actual question. This so-called query rewriting can help to create a more comprehensive version rather than a very narrow question. In the webinar we discussed, for example, the technique called Hypothetical Document Embedding. Here, a hypothetical answer is generated from the question, which is then vectorised in order to better match the existing documents. At the final output stage, the user naturally receives the answer to their original question; only in the background do we use a rewritten version for the database search.",[19,1272,1274],{"id":1273},"the-retrieval-pipeline","The Retrieval Pipeline",[11,1276,1277],{},"The final step decides how the actual answer is generated. The Retrieval Pipeline searches the knowledge database for the appropriate text passages by comparing the vector of the user's question with the vectors of the chunks. The most similar chunks are initially selected because they match best in terms of content. If there are too many possible hits, the list can be refined further using so-called re-ranking methods. A language model then evaluates, for example, which chunks are truly most relevant and reorders them.",[27,1279,1280],{},[11,1281,1282],{},"\"Retrieval means finding the right text passages using a query in the vector space by determining the semantic distance and filtering out which ones are closest in meaning.\"",[11,1284,1285],{},"After retrieval, the system can limit the volume of incoming text pieces to avoid overloading the generative language model with too much context. Too few chunks are equally problematic, however, because relevant information may then be missing. In the end all selected chunks are sent together with the original question and a system prompt to the large language model. These three elements, system prompt, context and query, ultimately result in a coherent text as the response.",[19,1287,1289],{"id":1288},"hallucination-and-sources-of-error","Hallucination and Sources of Error",[11,1291,1292],{},"In the webinar we were often asked whether a RAG system can hallucinate. Strictly speaking, any language generator can invent facts or mix up details when it lacks the right context. RAG does, however, reduce the likelihood of such errors. The AI bases its answers on external and generally verified content from the knowledge database. At the same time, it can still draw on its general language knowledge, which in rare cases can lead to incorrect additions. The key advantage is that the model works with queryable sources rather than spinning everything out of its own \"memory\".",[11,1294,1295],{},"When a document is very extensive, another problem can arise: the model receives only excerpts. Anyone expecting a global summary of the entire content runs up against the context limit. Planning a RAG system therefore requires thinking about whether a user needs the complete document or only individual details. Modern language models offer longer context windows, which partially resolves this problem and allows more extensive analyses when needed.",[19,1297,1299],{"id":1298},"conclusion-and-outlook","Conclusion and Outlook",[11,1301,1302],{},"My summary of the webinar showed how Retrieval Augmented Generation works in practice. It is not only a clever method for making vast amounts of data manageable; it is also an opportunity to sustainably improve internal research within organisations. Once you have experienced how quickly an AI system delivers a precise answer, you will rarely go back to rigid file structures, lengthy full-text searches or manual scrolling.",[11,1304,1305],{},"Setting up a RAG system does require expertise. The correct configuration of the ingestion pipeline, the integration of access permissions and metadata and the fine-tuning of query and retrieval often impose special requirements. In the next webinar we will focus more on the business-oriented side of generative AI solutions. We want to answer questions such as: \"What distinguishes Microsoft Copilot from ChatGPT?\" and \"When do you additionally need an AI solution like the AI Hub?\" This is intended to provide orientation for those who have to make decisions about introducing AI systems without needing to know every technical detail.",[27,1307,1308],{},[11,1309,1310],{},"\"Once you have experienced how efficiently a RAG system delivers precise answers, you will not want to return to traditional research methods.\"",[11,1312,1313],{},"With that, the webinar concluded and the most important questions about RAG were addressed. I thank everyone who took part and look forward to the next time, when we will go deeper into business perspectives and strategic considerations.",{"title":124,"searchDepth":125,"depth":125,"links":1315},[1316,1317,1318,1319,1320,1321],{"id":1232,"depth":128,"text":1233},{"id":1242,"depth":128,"text":1243},{"id":1263,"depth":128,"text":1264},{"id":1273,"depth":128,"text":1274},{"id":1288,"depth":128,"text":1289},{"id":1298,"depth":128,"text":1299},"RAG (Retrieval Augmented Generation) connects AI models with company-specific data to generate precise answers\nfrom extensive sources. The underlying pipeline processes documents in a structured way, recognises user\nintent and extracts relevant text passages from a knowledge database. Sensible splitting of texts, the\nmanagement of metadata and the avoidance of hallucinations all play a central role.\n","2024-07-03","52 Min.","bbv KI Webinar - Agent-RAG","\u002Fimages\u002Fpresentations\u002Fwebinar4\u002Frag_agent.png",{},"\u002Fpresentations\u002Fen\u002Fwebinar_agent_rag",{"title":1213,"description":1322},"presentations\u002Fen\u002Fwebinar_agent_rag","From Document to Knowledge","Agent-RAG","sGBbpHa7U5k","295","UeF_7zMlYGTUH7Pm9_FVJ6pTNajiJBdB8VL-t0hetSs",{"id":1337,"title":1338,"audience":1339,"body":1340,"carouselItems":136,"companyName":136,"date":136,"description":1546,"end_date":136,"eventDate":1547,"eventDuration":1548,"eventName":1549,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":1550,"mainTitle":136,"meta":1551,"navigation":144,"path":1552,"podcastId":136,"role":136,"seo":1553,"series":136,"start_date":136,"stats":136,"stem":1554,"subtitle":1555,"tags":136,"techs":136,"titleHighlight":1556,"videoId":1557,"views":1558,"watchTime":136,"__hash__":1559},"presentations\u002Fpresentations\u002Fen\u002Fwebinar_ai_in_2025.md","Webinar: Generative AI and Autonomous Agents","ca. 120 participants",{"type":8,"value":1341,"toc":1526},[1342,1347,1350,1352,1356,1359,1362,1366,1369,1374,1377,1380,1384,1387,1392,1395,1398,1402,1405,1408,1412,1415,1420,1423,1427,1430,1433,1437,1440,1445,1448,1452,1456,1459,1464,1467,1471,1474,1479,1482,1486,1489,1492,1496,1499,1503,1506,1510,1513,1516,1520,1523],[1343,1344,1346],"h2",{"id":1345},"review-and-outlook-making-sense-of-ai-in-2024-anticipating-2025","Review and Outlook: Making Sense of AI in 2024, Anticipating 2025",[11,1348,1349],{},"In public discourse, AI is often treated as if it consists of ChatGPT and a few image generators. That understates things, yet the observation says a great deal about the dynamics of 2024: language models are becoming synonymous with \"AI\" and thereby dominate both perception and investment decisions. That is exactly where this review starts. Anyone developing an AI strategy for an organisation needs less to know every individual headline and more to understand the big movements: which model category prevails? Who shapes the market? Which patterns are establishing themselves in usage, and which assumptions about progress and costs are realistic?",[15,1351],{":video-id":17},[19,1353,1355],{"id":1354},"_2024-as-the-year-of-language-models-and-why-that-matters-for-enterprises","2024 as the Year of Language Models - and Why That Matters for Enterprises",[11,1357,1358],{},"2024 is once again defined by Large Language Models (LLMs). This is not merely a continuation of 2023 but a shift in breadth: language models are no longer understood only as \"chat\" but as a generic capability layer. Organisations see that from the same basic capabilities (understanding, structuring, generating) a remarkable variety of business applications can be derived, from text work and research through to assistance in specialist processes.",[11,1360,1361],{},"At the same time the boundary between \"pure language model\" and multimodal AI is blurring. Image and audio capabilities are increasingly being integrated into the large models, making it appear as if everything is \"one model\" even though different modalities and product capabilities underlie it. For strategy this means: not every AI topic needs to be pursued in parallel. Anyone who consistently observes LLMs and their productisation already understands a large part of what moves the market. Other AI breakthroughs, for example in specialised domains such as protein folding, remain relevant but are often less directly transferable to general business processes.",[19,1363,1365],{"id":1364},"openai-as-the-discourse-driver-even-if-best-models-is-not-the-whole-truth","OpenAI as the Discourse Driver - Even If \"Best Models\" Is Not the Whole Truth",[11,1367,1368],{},"A second dominant pattern in 2024 is the strong centralisation of attention: OpenAI shapes the narrative around generative AI. This is evident less in the fact that OpenAI is unchallenged \"number one\" in every category, and more in the fact that product announcements, releases and terminology have such strong pull that other developments are overshadowed in perception. Even where competitors catch up technically or pull ahead in specific areas, public attention often remains with OpenAI.",[27,1370,1371],{},[11,1372,1373],{},"\"You can really just watch what OpenAI does and get a reasonably good sense of what is happening generally in the generative AI market.\"",[11,1375,1376],{},"Multimodality is a central leitmotif here: models that understand image input, handle speech as input and output and are adding video as a capability layer on the horizon. The strategic question for organisations is less \"which demo is impressive?\" and more: which capabilities are becoming stable and affordable enough to migrate into products and processes as building blocks?",[11,1378,1379],{},"In the video space 2024 also shows an interesting paradox: OpenAI dominates the headlines with \"Sora\", but the actual innovation front is broader. Video models existed before, and there are models that excel in particular qualities (for example photorealism). For practice this means: market observation must not run on brand names alone but must compare capabilities, quality criteria and integration options.",[19,1381,1383],{"id":1382},"assistants-copilots-and-the-pattern-humans-stay-in-the-drivers-seat","Assistants, Copilots and the Pattern \"Humans Stay in the Driver's Seat\"",[11,1385,1386],{},"At the application level, 2024 establishes \"assistant thinking\". Organisations often introduce generative AI in a way that gives employees a tool to consult on a point-by-point basis: draft an email, summarise text, structure ideas, accelerate small sub-tasks. This applies both to general tools (ChatGPT, etc.) and to product-bound copilots (for example in office or coding environments).",[27,1388,1389],{},[11,1390,1391],{},"\"The common denominator: the human sits in the driver's seat. The assistant waits until asked.\"",[11,1393,1394],{},"This pattern is pragmatic, quick to introduce and often the first step. At the same time it has side effects: it delegates the actual usage strategy to every individual. Employees must figure out for themselves when the use is worthwhile, how to prompt cleanly, which data may be shared, how to verify results, how to avoid plagiarism and where the limits lie. Without an accompanying enablement structure, a patchwork emerges: some use it effectively, others not at all or incorrectly, and many stay with the same \"low-hanging fruit\" as always.",[11,1396,1397],{},"A further practical problem concerns differentiation: from the user's perspective a chat window is a chat window. Whether company knowledge, RAG, additional tools or policies sit behind it is not automatically visible. This leads to false expectations (\"why can't the assistant do this as well as ChatGPT?\") or, conversely, risky behaviour (\"is this now internally safe, or public?\"). Anyone who seriously wants to anchor assistants in organisations therefore needs not only technology but product thinking: clear positioning, clear boundaries and clear communication.",[19,1399,1401],{"id":1400},"rag-becomes-state-of-the-art-and-at-the-same-time-less-dramatic-than-before","RAG Becomes \"State of the Art\" - and at the Same Time Less Dramatic Than Before",[11,1403,1404],{},"Retrieval Augmented Generation (RAG) stabilises in 2024 as the standard pattern. For many business-facing assistants it is the key: rather than hoping the model has \"learned\" the specialist knowledge \"somewhere\", deliberately provide relevant documents, guidelines, knowledge articles or process descriptions and generate answers from them. This gives considerably better control over timeliness, traceability and context fidelity.",[11,1406,1407],{},"At the same time, RAG design is shifting as context windows grow: when models can process very large contexts, \"the perfect selection of the three best snippets\" becomes less central. Instead the emphasis shifts toward robust retrieval strategies, data quality, permissions and a clean handover into context, along with the question of how much information is useful without diluting the answer.",[19,1409,1411],{"id":1410},"ai-becomes-dramatically-cheaper-a-strategy-factor-not-a-side-detail","AI Becomes Dramatically Cheaper - a Strategy Factor, Not a Side Detail",[11,1413,1414],{},"A core finding of 2024 is the price collapse: for comparable benchmark performance, costs are falling by orders of magnitude over just a few years. This fundamentally changes the economics of many use cases. Things that seem \"too expensive today\" can quickly slide into the range of \"running operating costs are negligible\". This creates new categories of applications: more volume, more automation, more continuous analysis, more \"always-on\" checks.",[27,1416,1417],{},[11,1418,1419],{},"\"Intelligence is getting cheaper. That opens up use cases that are currently still at the margin.\"",[11,1421,1422],{},"Importantly, not every task needs the \"strongest\" model. 2024 makes it clearer that model selection is an architecture decision. There are tasks where a smaller or cheaper model is perfectly adequate, while complex tasks are handled by targeted use of more powerful models. This differentiation is often a quick lever in organisations to reduce costs while enabling more applications.",[19,1424,1426],{"id":1425},"at-the-same-time-the-feeling-of-quantum-leaps-is-stagnating","At the Same Time, the Feeling of \"Quantum Leaps\" Is Stagnating",[11,1428,1429],{},"Parallel to the price collapse, 2024 presents a different picture on the pure \"intelligence curve\" of closed-source top models: the dramatic jumps perceived when moving from earlier model generations are flattening out. Rankings change, individual providers overtake each other in specific areas, but the steps feel incremental. This leads to a strategic correction: progress is not automatically extrapolatable in a linear fashion.",[11,1431,1432],{},"Several possible limiting factors underlie this: high-quality training data is not available in unlimited supply, scaling runs into technical and economic limits, and major breakthroughs often require new research approaches (architecture, training, data, optimisation). In practice this means: organisations should not wait for everything to be \"magically twice as good in six months\". The better approach is to amortise the capabilities of the current generation while observing the next wave in parallel.",[19,1434,1436],{"id":1435},"open-source-grows-up-privacy-vs-performance-is-no-longer-a-hard-choice","Open Source Grows Up - \"Privacy vs. Performance\" Is No Longer a Hard Choice",[11,1438,1439],{},"2024 is also a strong year for open-source models. The central change: open source is reaching a proximity to top models in many benchmarks that was previously not realistic. This shifts the classic dichotomy (\"closed source = good, open source = weak but private\"). Open-source models increasingly offer a combination of solid performance, low cost and high control.",[27,1441,1442],{},[11,1443,1444],{},"\"The switching costs are almost zero: a different endpoint, and suddenly massively lower costs.\"",[11,1446,1447],{},"For organisations this is strategically relevant for two reasons: first as a cost lever (depending on setup and hosting), second as a dependency question. When alternatives become realistic, the risk of being fully dependent on single providers or pricing models decreases. At the same time it remains important to think through the governance question: where do the models come from, what biases do they carry, which licences apply and how can this be used responsibly in a business context?",[1343,1449,1451],{"id":1450},"looking-into-the-crystal-ball-what-could-shape-2025","Looking into the Crystal Ball: What Could Shape 2025",[19,1453,1455],{"id":1454},"computer-usage-as-a-bridge-models-operating-interfaces-like-a-human","\"Computer Usage\" as a Bridge: Models Operating Interfaces Like a Human",[11,1457,1458],{},"One obvious trend is that models not only generate text but \"see\" the computer and operate it via UI actions: moving the mouse, clicking buttons, filling in fields, typing text. This looks like an interim solution but a very practical one: instead of building clean integrations and APIs everywhere, you use the interface that already exists. Particularly for software that is not AI-ready, this can enable quick automation.",[27,1460,1461],{},[11,1462,1463],{},"\"It is like using a sledgehammer on a nail, but a cool transitional phase.\"",[11,1465,1466],{},"Long-term, UI automation remains less elegant than direct interfaces. Short-term, however, it can close a gap: bridging processes, connecting tools, making legacy systems usable. Strategically this is less an \"end state\" than an accelerator for getting agentic workflows into real enterprise environments in the first place.",[19,1468,1470],{"id":1469},"reasoning-models-as-a-new-era-with-a-new-pricetime-profile","Reasoning Models as a New Era - with a New Price\u002FTime Profile",[11,1472,1473],{},"Probably the strongest shift in 2025 is the \"reasoning era\": models that do not answer immediately but think internally for longer, examine intermediate paths and generate step sequences before committing. The effect: on hard tasks, the hit rate rises significantly, but the price is not only monetary; it also shows up in latency and planability.",[27,1475,1476],{},[11,1477,1478],{},"\"The longer the model thinks, the better the performance, but under different conditions.\"",[11,1480,1481],{},"A decisive point: more performance costs disproportionately more. The last few percentage points of accuracy can become exponentially more expensive. This changes the decision matrix in organisations. The question is no longer only \"which model is best?\" but: how much accuracy is necessary for this use case? What does an additional quality level cost me, and is it economically justified? In many cases the pragmatic optimum lies not at maximum reasoning but at a well-balanced setup of cheaper base models, targeted context (RAG) and reasoning only where it truly counts.",[19,1483,1485],{"id":1484},"from-assistants-to-agents-virtual-employees-rather-than-passive-helpers","From Assistants to Agents: Virtual Employees Rather Than Passive Helpers",[11,1487,1488],{},"If 2024 keeps the human in the driver's seat, 2025 shifts the focus toward agents: systems that pursue goals, plan chains of tasks, execute intermediate steps autonomously and involve humans only where approvals, decisions or context are needed. This is a role reversal: the human assists the agent selectively rather than the other way round.",[11,1490,1491],{},"At the same time, 2024 shows why agents do not \"just work\": an open task, tools, data access and then \"go ahead\" is often unreliable. Agents lose focus, run into dead ends or burn budget without results. 2025 therefore sees a methodological shift: workflow-based agentics. First guide tightly, then grant freedoms incrementally. This creates not only more reliability but also better engineering: agents become steerable process components rather than unpredictable demos.",[19,1493,1495],{"id":1494},"agentic-process-automation-integrating-processes-and-then-rethinking-them","Agentic Process Automation: Integrating Processes - and Then Rethinking Them",[11,1497,1498],{},"The next step after \"agent in an existing process\" is redesigning the processes themselves. Agents are different from humans: less deeply specialised, but broadly deployable, quick to switch context, strong in text\u002Fanalysis\u002Fstructure and with access to tools. This can cut across silos. Processes that were previously strictly distributed across departments can in future be structured differently: more end-to-end, more data-driven, more organised around capability bundles rather than role profiles. 2025 is more the year of experiments and discussions than of widespread implementation, but precisely these discussions are often the beginning of genuine productivity gains.",[19,1500,1502],{"id":1501},"china-as-accelerator-open-source-as-a-geopolitical-variable","China as Accelerator - Open Source as a Geopolitical Variable",[11,1504,1505],{},"A further trend is the growing role of Chinese models, particularly in the open-source world. If capable models from China come to define the open-source standard, this will reinforce not only technical questions but also governance and values questions: cultural biases, content guardrails, training data, political frameworks. At the same time it holds true that biases and interests exist in US models as well. The shift makes the topic more visible and forces organisations to decide more consciously which dependencies they are willing to enter.",[19,1507,1509],{"id":1508},"the-agi-debate-gets-louder-and-remains-hard-to-grasp","The AGI Debate Gets Louder - and Remains Hard to Grasp",[11,1511,1512],{},"2025 is likely to be \"AGI-heavy\" also because the term serves as a marker for progress and market leadership. At the same time AGI remains vaguely defined. Even if a model reaches \"human level\" on many benchmarks, this does not automatically mean it immediately delivers massive business value in practice. Reasoning can be expensive, slow and genuinely superior only on certain types of tasks.",[11,1514,1515],{},"It is also relevant that economic and strategic interests can play a role in communications, for example through contractual constructs where declaring \"AGI\" could influence certain partnerships or conditions. For organisations the conclusion is a sober stance: observe the AGI debate, but weight use cases, cost profiles, integration capability and risk assessment more highly than labels.",[1343,1517,1519],{"id":1518},"concluding-line-for-strategy","Concluding Line for Strategy",[11,1521,1522],{},"2024 shows an operational reality: language models are the dominant capability layer, OpenAI shapes the discourse, assistants are the most widespread entry point, RAG is becoming standard and costs are falling dramatically. At the same time, the impression of \"intelligence quantum leaps\" in classic models is flattening, while open source is maturing and creating genuine alternatives.",[11,1524,1525],{},"2025 feels like a transition into a new era: more agentics, more process integration, more reasoning, but under new conditions: higher costs for top performance, more latency, more engineering discipline, more governance questions. The pragmatic consequence is two-pronged: consistently make the capabilities of the current generation productive and amortise them, and observe the next wave closely enough to be prepared when price and maturity tip.",{"title":124,"searchDepth":125,"depth":125,"links":1527},[1528,1537,1545],{"id":1345,"depth":125,"text":1346,"children":1529},[1530,1531,1532,1533,1534,1535,1536],{"id":1354,"depth":128,"text":1355},{"id":1364,"depth":128,"text":1365},{"id":1382,"depth":128,"text":1383},{"id":1400,"depth":128,"text":1401},{"id":1410,"depth":128,"text":1411},{"id":1425,"depth":128,"text":1426},{"id":1435,"depth":128,"text":1436},{"id":1450,"depth":125,"text":1451,"children":1538},[1539,1540,1541,1542,1543,1544],{"id":1454,"depth":128,"text":1455},{"id":1469,"depth":128,"text":1470},{"id":1484,"depth":128,"text":1485},{"id":1494,"depth":128,"text":1495},{"id":1501,"depth":128,"text":1502},{"id":1508,"depth":128,"text":1509},{"id":1518,"depth":125,"text":1519},"The webinar puts the most important developments in the AI year 2024 into context and shows why language\nmodels are shaping the market and public discourse. One focus is on OpenAI as the dominant pace-setter and\non the trend toward assistants and copilots in enterprises, including typical adoption and differentiation\nquestions. It also covers how sharply AI usage costs are falling and why the perceived intelligence jumps\nin leading models are simultaneously getting smaller. A further section examines the rise of capable\nopen-source models and their relevance for costs, control and dependency. The outlook for 2025 centres on\ncomputer-assisted operation, reasoning models with new cost profiles, agentic workflows and the growing role\nof Chinese models alongside the increasingly loud AGI debate.\n","2025-01-22","88 Min.","bbv KI Webinar - Review\u002FOutlook 2024\u002F2025","\u002Fimages\u002Fpresentations\u002Fwebinar6\u002Fai_in_2025_1_1200x1200.png",{},"\u002Fpresentations\u002Fen\u002Fwebinar_ai_in_2025",{"title":1338,"description":1546},"presentations\u002Fen\u002Fwebinar_ai_in_2025","The Most Important Developments of 2024 and What 2025 Could Bring","Review\u002FOutlook 2024\u002F2025","sfOCY-lSGRg","484","-clBMVFCL56HxdrzhkCCsoMVl05XDAxjEhOvjHYB_v0",{"id":1561,"title":1562,"audience":1563,"body":1564,"carouselItems":136,"companyName":136,"date":136,"description":1737,"end_date":136,"eventDate":1738,"eventDuration":1739,"eventName":1740,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":1741,"mainTitle":136,"meta":1742,"navigation":144,"path":1743,"podcastId":136,"role":136,"seo":1744,"series":136,"start_date":136,"stats":136,"stem":1745,"subtitle":1746,"tags":136,"techs":136,"titleHighlight":1747,"videoId":1748,"views":1749,"watchTime":136,"__hash__":1750},"presentations\u002Fpresentations\u002Fen\u002Fwebinar_ai_in_2026.md","Webinar: AI Trends 2026","ca. 200 participants",{"type":8,"value":1565,"toc":1723},[1566,1570,1573,1575,1579,1582,1587,1590,1594,1597,1602,1605,1609,1612,1617,1620,1624,1627,1632,1635,1639,1642,1645,1649,1652,1655,1659,1662,1667,1670,1674,1677,1680,1684,1687,1708,1711,1715,1718],[1343,1567,1569],{"id":1568},"review-2025-and-outlook-2026-ai-between-convergence-platforms-and-sovereignty","Review 2025 and Outlook 2026: AI Between Convergence, Platforms and Sovereignty",[11,1571,1572],{},"The review of 2025 serves not chronology but orientation: which forces are actually shaping the market, and which developments are plausible for 2026? Rather than individual product announcements, the focus is on recurring patterns: how is the performance dynamic of models changing? Where does differentiation take place? What does this mean for costs, integration and dependencies in organisations?",[15,1574],{":video-id":17},[19,1576,1578],{"id":1577},"_2025-as-the-year-of-model-convergence","2025 as the Year of Model Convergence",[11,1580,1581],{},"At first glance 2025 looks less spectacular because classic benchmarks barely show major jumps any more. This is less a sign of stagnation and more an indication that many established measuring sticks have been \"maxed out\". Progress is shifting: not breaking the next percentage points dominates, but efficiency, smaller models, lower resource requirements, stronger performance per compute unit and increasingly smartphone-capable variants.",[27,1583,1584],{},[11,1585,1586],{},"\"The challenge is no longer to crack the benchmark, but to reach it with ever smaller, more efficient models.\"",[11,1588,1589],{},"At the same time a subjective effect emerges: for everyday tasks, progress often feels like stagnation because many standard tasks are already \"good enough\" solved. Measurably, capabilities continue to rise, but increasingly in niches that only become noticeable on domain-specific tasks.",[19,1591,1593],{"id":1592},"business-perspective-model-choice-becomes-secondary-platform-choice-central","Business Perspective: Model Choice Becomes Secondary, Platform Choice Central",[11,1595,1596],{},"When models converge, the question \"which model is the most intelligent?\" loses significance. More relevant criteria become context length, cost structure, availability, compliance, data flow and above all integration capability. Differentiation thereby shifts clearly away from the model core toward the platform: interface, search and retrieval components, memory, vector databases, governance, transparency and agentic workflows.",[27,1598,1599],{},[11,1600,1601],{},"\"The great differentiator is no longer the model; it is the platform around it.\"",[11,1603,1604],{},"Many functions that in daily use feel like \"model capabilities\" (web search, source display, document retrieval, memory) are in practice platform functions. This means: perceived progress increasingly arises from productisation and ecosystem rather than from raw model intelligence.",[19,1606,1608],{"id":1607},"usa-vs-china-infrastructure-bet-against-open-weight-strategy","USA vs. China: Infrastructure Bet Against Open-Weight Strategy",[11,1610,1611],{},"A second leitmotif of 2025 is the rivalry between the USA and China, visible through two contrasting strategies. In the USA enormous infrastructure investments dominate: data centres, training and inference, with the bet that more compute leads to better market position. China relies more on open weights: models and research are made broadly available, accelerating innovation at scale and generating price and efficiency pressure.",[27,1613,1614],{},[11,1615,1616],{},"\"Intelligence is getting cheaper, not necessarily because everything becomes cheaper, but because performance per dollar is rising strongly.\"",[11,1618,1619],{},"This dynamic lowers costs for all market participants, accelerates open-source ecosystems and simultaneously shifts scientific influence. For organisations this means: options are becoming more diverse, and architecture decisions are gaining ground over provider loyalty.",[19,1621,1623],{"id":1622},"_2025-promises-are-technically-redeemed-and-generate-ai-slop","2025: Promises Are Technically Redeemed - and Generate \"AI Slop\"",[11,1625,1626],{},"In several modalities 2025 feels like a year of maturity: text intelligence, large contexts, image generation and increasingly video deliver results that in many cases no longer stumble against clear technical limits. At the same time, the volume of generated content is rising faster than its quality. This produces \"AI slop\": content produced without care and published without verification, with the risk of feedback loops in which AI references other AI content as seemingly legitimate sources.",[27,1628,1629],{},[11,1630,1631],{},"\"The thinking belongs to people; the AI can do the work.\"",[11,1633,1634],{},"The same mechanism shows up in code: the bottleneck shifts from writing to reviewing, securing and classifying. When a great deal of code emerges very quickly, the risk of unverified adoption and security gaps grows, even when the underlying technology is impressive.",[19,1636,1638],{"id":1637},"outlook-2026-ai-value-arises-through-process-change-not-through-the-next-model","Outlook 2026: AI Value Arises Through Process Change, Not Through the Next Model",[11,1640,1641],{},"For 2026 what is taking shape is less \"new model magic\" and more an implementation question: the capabilities are there. What will be decisive is whether organisations are willing to adapt their working methods and processes. Experience from software development shows a typical pattern: in the early adoption phase a productivity trough first arises (tool selection, new workflows, new roles), before significant acceleration becomes possible with consistent adaptation.",[11,1643,1644],{},"What is essential here is a shift in role: less \"writing code\" or \"generating output\", more \"specifying requirements\", \"providing context\", \"reviewing results\" and \"structuring systems so that agents can work reliably\".",[19,1646,1648],{"id":1647},"_2026-as-the-year-of-integration","2026 as the Year of Integration",[11,1650,1651],{},"In the private sphere AI often feels frictionless because users are already working in integrated ecosystems. In organisations that is rarely the case: CRM, ERP, databases, specialist applications, legacy systems and individual process landscapes cannot simply be transferred into a single provider ecosystem. Integration therefore becomes the core condition for AI to genuinely create value in business.",[11,1653,1654],{},"Without integration, AI remains \"a chat window next to reality\". With integration it becomes part of workflows: data access, tool access, permissions, auditability, approvals and traceable intermediate steps.",[19,1656,1658],{"id":1657},"mcp-as-a-standard-connecting-ai-platforms-and-business-tools","MCP as a Standard: Connecting AI Platforms and Business Tools",[11,1660,1661],{},"The Model Context Protocol (MCP) is establishing itself as the mechanism for connecting AI clients (platforms, IDEs, interfaces) with MCP servers (tools, data sources, business systems). The difference from classic APIs lies not in the existence of an interface but in the fact that the interface is conceived agentically: capabilities are describable and usable by models, including interaction patterns such as clarifying questions or human-in-the-loop.",[27,1663,1664],{},[11,1665,1666],{},"\"Not every organisation has to build the interface. An organisation can also deliver value purely as an MCP server.\"",[11,1668,1669],{},"This also changes business models: value can be created by offering capabilities in a way that allows users to consume them from their preferred AI platform, rather than necessarily pulling users into a proprietary UI.",[19,1671,1673],{"id":1672},"_2026-as-the-year-of-the-login-lock-in-arises-where-investment-is-made","2026 as the Year of the Login: Lock-In Arises Where Investment Is Made",[11,1675,1676],{},"Because models are becoming more interchangeable, lock-in is shifting to the platform and implementation level: chat history, memory, audit logs and above all the investments in data projects, curated knowledge bases, agent workflows, governance and integrations. The more that is \"built\" in a platform, the higher the switching costs become.",[11,1678,1679],{},"The greatest lock-in arises not through the model but through your own work that has been poured into platform and process structures.",[19,1681,1683],{"id":1682},"sovereignty-on-three-levels-data-centre-platform-business-integration","Sovereignty on Three Levels: Data Centre, Platform, Business Integration",[11,1685,1686],{},"Sovereignty cannot be reduced to hosting. Relevant decisions concern at least three levels:",[1688,1689,1690,1696,1702],"ol",{},[342,1691,1692,1695],{},[745,1693,1694],{},"Data centre \u002F operations:"," where do models run, where does data reside physically and legally?",[342,1697,1698,1701],{},[745,1699,1700],{},"AI platform:"," who controls memory, audit logs, governance, agentics and extensibility?",[342,1703,1704,1707],{},[745,1705,1706],{},"Business integration:"," which systems are connected, which data flows arise, how controllable are they?",[11,1709,1710],{},"Open-source options and local providers can play a role at every level. What is decisive is that requirements (data protection, sector regulations, operating model, transparency, extensibility) are made explicit and implemented consistently across all levels.",[19,1712,1714],{"id":1713},"concluding-line-for-2026","Concluding Line for 2026",[11,1716,1717],{},"2026 will be interesting less because of the \"next model bang\" and more through implementation: integration, clean data work, agentic workflows with governance and the willingness to adapt processes and role definitions. The technology provides the building blocks. Value arises where those building blocks are embedded in the system landscape and in everyday operations.",[27,1719,1720],{},[11,1721,1722],{},"\"Don't be dazzled: the thinking belongs to people; the AI can do the work.\"",{"title":124,"searchDepth":125,"depth":125,"links":1724},[1725],{"id":1568,"depth":125,"text":1569,"children":1726},[1727,1728,1729,1730,1731,1732,1733,1734,1735,1736],{"id":1577,"depth":128,"text":1578},{"id":1592,"depth":128,"text":1593},{"id":1607,"depth":128,"text":1608},{"id":1622,"depth":128,"text":1623},{"id":1637,"depth":128,"text":1638},{"id":1647,"depth":128,"text":1648},{"id":1657,"depth":128,"text":1658},{"id":1672,"depth":128,"text":1673},{"id":1682,"depth":128,"text":1683},{"id":1713,"depth":128,"text":1714},"The review of 2025 frames the development of AI models as a phase of convergence in which progress is\nvisible less through classic benchmarks and more through efficiency and cost per performance. One focus is\non the shift in differentiation from model quality to platform functions such as integration, memory,\nretrieval, governance and agentic workflows. The rivalry between the USA and China is framed as an\ninterplay of massive infrastructure investments and an open-weight strategy with strong price pressure and\na growing open-source ecosystem. For 2026, integration into existing enterprise systems and standards such\nas MCP are central, to connect AI workflows reliably with business applications. At the same time, lock-in\neffects and AI sovereignty are presented as decisions about data centre, platform and connected systems.\n","2026-01-14","107 Min.","bbv KI Webinar - Review\u002FOutlook 2025\u002F2026","\u002Fimages\u002Fpresentations\u002Fwebinar9\u002Fai_trends_2026_1",{},"\u002Fpresentations\u002Fen\u002Fwebinar_ai_in_2026",{"title":1562,"description":1737},"presentations\u002Fen\u002Fwebinar_ai_in_2026","AI Wrapped 2025 and Outlook 2026","Review\u002FOutlook 2025\u002F2026","5kYoGZ2qXWc","2224","88Ehawynj0UjdS5eKt2pEpct10Plb4B-zngLosWa-8Y",{"id":1752,"title":1753,"audience":1754,"body":1755,"carouselItems":136,"companyName":136,"date":136,"description":1891,"end_date":136,"eventDate":1892,"eventDuration":1893,"eventName":1894,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":1895,"mainTitle":136,"meta":1896,"navigation":144,"path":1897,"podcastId":136,"role":136,"seo":1898,"series":136,"start_date":136,"stats":136,"stem":1899,"subtitle":1900,"tags":136,"techs":136,"titleHighlight":1901,"videoId":1902,"views":1903,"watchTime":136,"__hash__":1904},"presentations\u002Fpresentations\u002Fen\u002Fwebinar_ai_infra.md","Webinar: Robust AI Foundation","ca. 30 participants",{"type":8,"value":1756,"toc":1879},[1757,1760,1762,1766,1769,1774,1777,1781,1784,1789,1792,1796,1799,1803,1806,1809,1813,1816,1819,1827,1831,1834,1837,1841,1844,1847,1851,1854,1859,1862,1866,1869,1873,1876],[11,1758,1759],{},"As part of the bbv webinar series on artificial intelligence, this episode from 20 November 2025 focuses on AI infrastructure as the foundation for modern applications. The session is moderated by Emre; I contribute practical insights into which building blocks organisations need for stable and scalable AI operations. The focus is less on individual AI use cases and more on the reusable platform components that enable model changes, cost control, data integration, quality assurance and traceability. The goal is a sober assessment of how AI infrastructure, as an organisational and technical foundation, supports long-term viable AI solutions.",[15,1761],{":video-id":17},[19,1763,1765],{"id":1764},"ai-infrastructure-as-the-foundation-for-scalable-applications","AI Infrastructure as the Foundation for Scalable Applications",[11,1767,1768],{},"When the term \"AI infrastructure\" comes up in projects, many people first think of GPU clusters, compute power and the cloud. In practice, however, the bottleneck often lies elsewhere: AI only becomes sustainably usable when there is a cross-use-case foundation that solves recurring requirements centrally, regardless of whether the end result is a chatbot, an assistance system or a background agent.",[27,1770,1771],{},[11,1772,1773],{},"\"Today we are talking about the hob, the oven and that kind of thing.\"",[11,1775,1776],{},"Infrastructure here does not mean the end product but the base on which multiple AI applications can reliably be built. In the blueprint, AI infrastructure sits at the bottom as the foundation, with various use cases built on top and sharing that base.",[19,1778,1780],{"id":1779},"making-models-centrally-interchangeable","Making Models Centrally Interchangeable",[11,1782,1783],{},"A frequent requirement from organisations: models must be exchangeable quickly and centrally. The reasons are vendor lock-in, different model sizes (cost vs. quality), data protection requirements (for example location\u002Fjurisdiction) and the rapid development of new model generations.",[27,1785,1786],{},[11,1787,1788],{},"\"New models come along at a monthly pace. You naturally want to enable a fast switch.\"",[11,1790,1791],{},"Technically, switching within comparable model types is often feasible, but behaviour can change. An infrastructure that treats model changes as a controlled operational process is therefore required.",[19,1793,1795],{"id":1794},"lm-proxies-and-gateways","LM Proxies and Gateways",[11,1797,1798],{},"This is where LLM proxies and LLM gateways come in: use cases do not address specific provider models directly but stable internal endpoints such as \"thinking-large\" or \"embedding-small\". In the gateway the mapping to the model behind that endpoint is maintained. This allows central switching without adapting code in every use case, an adapter layer between application and model.",[19,1800,1802],{"id":1801},"keeping-costs-plannable-and-monitorable","Keeping Costs Plannable and Monitorable",[11,1804,1805],{},"With \"LLM as a service\", costs are usually token-based and therefore dynamic: depending on request volume, text lengths, tool calls and agent steps. Organisations therefore want to set budgets, see costs per team\u002Fkey\u002Fmodel and prevent misuse (for example with externally accessible bots).",[11,1807,1808],{},"Because requests run through the gateway, it can measure, break down and limit token consumption and costs (limits, rate limits, alerts). Centralised access management is often attached to this as well: API keys, roles, user groups and policies.",[19,1810,1812],{"id":1811},"integrating-company-knowledge-via-rag","Integrating Company Knowledge via RAG",[11,1814,1815],{},"Language models know public training knowledge but not an organisation's current internal knowledge. To make that knowledge usable, Retrieval Augmented Generation (RAG) is typically used: documents are split into sections, semantically indexed and, when a query arrives, relevant text passages are found and passed to the model in context.",[11,1817,1818],{},"This requires above all:",[339,1820,1821,1824],{},[342,1822,1823],{},"Embedding models for vectorisation",[342,1825,1826],{},"Vector databases for semantic search",[19,1828,1830],{"id":1829},"ingestion-keeping-data-current","Ingestion: Keeping Data Current",[11,1832,1833],{},"RAG only works when data is current and controlled. Documents change, must be deleted (compliance) or outdated versions should no longer appear. An ingestion pipeline therefore belongs to the infrastructure: it monitors data sources (for example SharePoint, Confluence, file shares), detects changes and updates the vector database.",[11,1835,1836],{},"A practical point: poor version management (\"v1\u002Fv2\u002Fv3\" in parallel) leads to duplicates and wastes context. It is better to archive old versions and index only current sources.",[19,1838,1840],{"id":1839},"assuring-quality-evaluation-rather-than-flying-blind","Assuring Quality: Evaluation Rather Than Flying Blind",[11,1842,1843],{},"When models can be exchanged centrally and systems change, evaluation becomes central: how do you detect whether a model change, prompt update or new data pipeline has improved or degraded quality?",[11,1845,1846],{},"A typical evaluation setup includes test questions, reference answers (and for RAG optionally expected sources) and metrics such as correctness, completeness and conciseness. LLM-as-a-judge is frequently used to generate evaluations automatically at scale. Ideally, evaluation happens before a switch (comparing old vs. new) and continuously during operations.",[19,1848,1850],{"id":1849},"traceability-through-observability","Traceability Through Observability",[11,1852,1853],{},"AI systems consist of multiple processing steps: context assembly, query rewriting, retrieval, guardrails, tool calls, post-processing. When something goes wrong, the decisive question is which step influenced the result.",[27,1855,1856],{},[11,1857,1858],{},"\"Every step should be traceable.\"",[11,1860,1861],{},"For this, logs and traces are used (for example OpenTelemetry, OpenInference) along with suitable observability tools that visualise requests as a flow. This is relevant both for debugging and for operations and governance.",[19,1863,1865],{"id":1864},"data-protection-pii-detection-as-the-last-line-of-defence","Data Protection: PII Detection as the Last Line of Defence",[11,1867,1868],{},"With external model services, the risk arises that requests are at least temporarily stored by the provider or processed outside the desired jurisdiction. A PII detection step is therefore often inserted as the last step before the model: personal data (name, email, IBAN etc.) is replaced or the request is blocked. Implemented centrally in the gateway, this applies automatically to all use cases.",[19,1870,1872],{"id":1871},"platform-approach-rather-than-individual-solutions","Platform Approach Rather Than Individual Solutions",[11,1874,1875],{},"The common denominator: many requirements repeat themselves across all AI applications. When each application solves them separately, inconsistencies arise in model access, cost control, data integration, quality measurement and observability. The platform approach bundles these topics as infrastructure building blocks so that new use cases can be built faster and operated consistently.",[11,1877,1878],{},"Within this framework I also position the bbv AI Hub: as an \"opinionated\" assembly of modules (for example gateway, RAG stack, ingestion, evaluation, observability, PII protection) plus integrations that are needed again and again in projects, not as a single feature but as a stable foundation for multiple AI solutions.",{"title":124,"searchDepth":125,"depth":125,"links":1880},[1881,1882,1883,1884,1885,1886,1887,1888,1889,1890],{"id":1764,"depth":128,"text":1765},{"id":1779,"depth":128,"text":1780},{"id":1794,"depth":128,"text":1795},{"id":1801,"depth":128,"text":1802},{"id":1811,"depth":128,"text":1812},{"id":1829,"depth":128,"text":1830},{"id":1839,"depth":128,"text":1840},{"id":1849,"depth":128,"text":1850},{"id":1864,"depth":128,"text":1865},{"id":1871,"depth":128,"text":1872},"The webinar focuses on AI infrastructure as the shared foundation on which multiple AI applications in\nan organisation can be built. Central components covered include LLM gateways for interchangeable model\nusage and mechanisms for cost and access control with token-based services. A further focus is access to\ncompany knowledge via RAG, vector databases and ingestion pipelines for continuously updating documents.\nTo safeguard quality, evaluation approaches are presented that make model changes and other modifications\nmeasurable. The session also covers observability and data protection, including traceability of processing\nsteps and PII detection before requests are sent to external models.\n","2025-11-20","51 Min.","bbv KI Webinar - Robust AI Foundation","\u002Fimages\u002Fpresentations\u002Fwebinar8\u002Fai_infra_1.png",{},"\u002Fpresentations\u002Fen\u002Fwebinar_ai_infra",{"title":1753,"description":1891},"presentations\u002Fen\u002Fwebinar_ai_infra","A Platform Approach for Long-Term Enterprise Success","Robust AI Foundation","UKnhaiKy1_w","67","C6L37LX9o57WWp8BfgVOcSa3vSURSZQ1L8X2zrWxSTw",{"id":1906,"title":1907,"audience":727,"body":1908,"carouselItems":136,"companyName":136,"date":136,"description":2026,"end_date":136,"eventDate":2027,"eventDuration":2028,"eventName":2029,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":2030,"mainTitle":136,"meta":2031,"navigation":144,"path":2032,"podcastId":136,"role":136,"seo":2033,"series":136,"start_date":136,"stats":136,"stem":2034,"subtitle":217,"tags":136,"techs":136,"titleHighlight":2035,"videoId":2036,"views":2037,"watchTime":136,"__hash__":2038},"presentations\u002Fpresentations\u002Fen\u002Fwebinar_datenflüsse.md","Webinar: Data Flows in AI Systems",{"type":8,"value":1909,"toc":2019},[1910,1913,1916,1921,1924,1926,1930,1933,1936,1941,1944,1948,1951,1954,1959,1962,1966,1969,1972,1977,1980,1984,1987,1990,1995,1998,2002,2005,2008,2013,2016],[11,1911,1912],{},"I held a webinar that took an intensive look at data flows in AI systems. The central question was how large language models, such as GPT, actually work and exactly where we as users have the decisive lever to guarantee the quality and security of responses. My colleague Allen moderated the evening, introduced our previous topics in the webinar series and emphasised from the start that we wanted to take \"a look under the hood\".",[11,1914,1915],{},"It was important to me to show participants that AI is not magic and does not generate knowledge \"out of thin air\" on its own. The technology is based on statistical methods and mathematical models that only prove helpful when we supply them with the right information. At the same time I wanted to make understandable why we should keep a close eye on so-called \"data flows\" as soon as we integrate AI models into our workflows.",[27,1917,1918],{},[11,1919,1920],{},"\"Generative AI is not a magic trick. It only works through interaction and the deliberate control of the underlying data.\"",[11,1922,1923],{},"This statement ran as a connecting thread through the entire webinar. I wanted to emphasise that every user directly influences the outcome the moment they ask questions or transmit information.",[15,1925],{":video-id":17},[19,1927,1929],{"id":1928},"three-essential-sources-for-the-model","Three Essential Sources for the Model",[11,1931,1932],{},"To clarify what is meant by \"data flows\", I explained which sources supply a large language model like GPT with information. First, there is the knowledge we provide directly when asking a question. Second, company-specific databases can be integrated to provide additional or current knowledge. And third, there is the so-called model knowledge, the statistical relationships that have become embedded in the model from its training data.",[11,1934,1935],{},"I described how important it is for many organisations to be able to connect their internal expertise, such as documents, manuals or guidelines, in the form of a database. Especially when the model's standard answer falls short or perhaps even contains outdated facts, this approach yields significantly higher accuracy and relevance. Otherwise the answers rest solely on the model's enormous but possibly no longer fully current or suitably oriented training material.",[27,1937,1938],{},[11,1939,1940],{},"\"The model brings its own 'general knowledge', but it only becomes truly precise when we make clear what is relevant to our specific case.\"",[11,1942,1943],{},"I noted that this second data source, company-specific knowledge, is our most important supplement to the model's more global knowledge. How best to integrate it depends, among other things, on the software architecture used and the data protection requirements.",[19,1945,1947],{"id":1946},"how-context-is-formed","How Context Is Formed",[11,1949,1950],{},"The webinar also covered how we can help the model formulate appropriate answers. I spoke about \"Retrieval-Augmented Generation\". Behind this term is the idea of not simply sending a question to the model but first deliberately retrieving information from a knowledge database and attaching it to the model as context. This way, not all possible documents are given to the system indiscriminately, only those that are truly relevant.",[11,1952,1953],{},"This approach reduces the risk of incorrect or contradictory statements. Above all it supports traceability. If the question arises afterwards as to why the AI reached a particular result, one can easily see which text passages served as the basis.",[27,1955,1956],{},[11,1957,1958],{},"\"Instead of confusing the model, we give it an exact selection of relevant content. This keeps the answer more precise and consistent.\"",[11,1960,1961],{},"I explained using an example that a language model responding to the query \"What does it mean when someone is on the bench?\" can go in very different directions. Is it thinking of a park bench, a financial institution, or the sporting context? Anyone who clarifies the appropriate background information from the outset will generally receive considerably more precise results.",[19,1963,1965],{"id":1964},"agents-as-supporting-actors","Agents as Supporting Actors",[11,1967,1968],{},"In the course of the webinar I showed that a third component can also come into play: software agents that take on different functions. One agent could be specialised in retrieving financial data, while another handles sales questions. These small helpers orchestrate how the large language model should respond to certain inputs.",[11,1970,1971],{},"Many participants were particularly surprised at this point by how many processes can be automated or predefined before the actual language model encounters a question. I described how this ensures that not every person has to type the perfect commands (\"prompts\") themselves, but that certain steps are automatically handled by agents.",[27,1973,1974],{},[11,1975,1976],{},"\"An agent can retrieve the passages, prepare them sensibly and if necessary go through several intermediate steps before querying the model.\"",[11,1978,1979],{},"My experience shows that this can be a blessing in complex environments such as large organisations. Users save time, get consistent results and do not have to play the entire keyboard of AI control manually every time.",[19,1981,1983],{"id":1982},"handling-sensitive-data","Handling Sensitive Data",[11,1985,1986],{},"With all the enthusiasm for the topic, limits and risks were also addressed. I made clear that a language model does not \"continue learning\" during use; it does not maintain a growing database in the background that automatically stores our inputs. Nevertheless, many providers store conversation histories to improve the system later or to train new models. This is a question of company policy and contractual arrangements.",[11,1988,1989],{},"An important point was the data protection perspective. Anyone who shares personal information with an external provider must carefully check which data is leaving the server and whether it should perhaps be anonymised. I explained that architectural solutions exist that process sensitive data in a separate environment before passing it on to the large model.",[27,1991,1992],{},[11,1993,1994],{},"\"Sensitive data should not simply be dumped into a publicly accessible AI system without thinking. Technically this can often be solved, but it must be approached consciously.\"",[11,1996,1997],{},"I made clear that an AI system like GPT does not guarantee error-free output. It can \"hallucinate\" when context is wrong or missing. A systematic approach to sources and a final check by humans remain essential in many cases. It was important to me to communicate this clearly so as not to raise false expectations.",[19,1999,2001],{"id":2000},"outlook-and-closing-thoughts","Outlook and Closing Thoughts",[11,2003,2004],{},"After illuminating the various aspects, from integrating a vector database through agent concepts to security, Allen concluded by previewing the next webinar in our series. On 11 April, Emre will guide us into the world of business application of AI. The focus will be on how to develop a viable AI product from a good idea and which strategies have proven themselves.",[11,2006,2007],{},"I personally look forward to this contribution greatly because it spans the arc from technology to real business practice. Often the technical feasibility is already there, but the value and successful implementation in everyday business life are challenges of their own.",[27,2009,2010],{},[11,2011,2012],{},"\"Successful AI projects emerge at the intersection of a good idea, efficient use of data and realistic implementation strategies.\"",[11,2014,2015],{},"At the end of the webinar there was a window for questions. It turned out that many participants apparently found the connections very clear, as no specific follow-up questions came. Allen joked that we were almost a little disappointed, but usually silence means the message was understood.",[11,2017,2018],{},"With that I closed the evening. My goal was to show how a language model really \"ticks\" in practice and why we should keep a close eye on all data flows. For me it is settled that a sound understanding of this process is the key to using AI solutions effectively and responsibly. I thank everyone who attended the webinar and look forward to continuing to develop this knowledge together.",{"title":124,"searchDepth":125,"depth":125,"links":2020},[2021,2022,2023,2024,2025],{"id":1928,"depth":128,"text":1929},{"id":1946,"depth":128,"text":1947},{"id":1964,"depth":128,"text":1965},{"id":1982,"depth":128,"text":1983},{"id":2000,"depth":128,"text":2001},"Large language models such as GPT are based on statistical relationships and the interaction between user,\ndatabase and model. The combination of general model knowledge and company-specific data sources improves\nthe precision and timeliness of answers. The focus is also on data security and the careful handling of\nconfidential information.\n","2024-02-28","41 Min.","bbv KI Webinar - Data Flows","\u002Fimages\u002Fpresentations\u002Fwebinar3\u002Fdatenfluesse.png",{},"\u002Fpresentations\u002Fen\u002Fwebinar_datenflusse",{"title":1907,"description":2026},"presentations\u002Fen\u002Fwebinar_datenflüsse","Data Flows","UMWRTP79vg8","275","mjn2-LviLuBqTxOrvzD654qqAvBqHSTuWB5MRpQWzDU",{"id":2040,"title":2041,"audience":1214,"body":2042,"carouselItems":136,"companyName":136,"date":136,"description":2189,"end_date":136,"eventDate":2190,"eventDuration":1324,"eventName":2191,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":2192,"mainTitle":136,"meta":2193,"navigation":144,"path":2194,"podcastId":136,"role":136,"seo":2195,"series":136,"start_date":136,"stats":136,"stem":2196,"subtitle":2197,"tags":136,"techs":136,"titleHighlight":2198,"videoId":2199,"views":2200,"watchTime":136,"__hash__":2201},"presentations\u002Fpresentations\u002Fen\u002Fwebinar_ki_agenten_in_der_praxis.md","Webinar: AI Agents in Practice",{"type":8,"value":2043,"toc":2179},[2044,2047,2052,2055,2057,2061,2064,2069,2072,2075,2080,2083,2087,2090,2093,2097,2100,2103,2108,2111,2114,2117,2120,2124,2127,2130,2134,2137,2140,2144,2147,2150,2154,2157,2160,2165,2168,2171,2173,2176],[11,2045,2046],{},"I spoke in a webinar together with my colleague Stefan Heberling on a topic that has fascinated me for a long time: generative AI and the so-called agentic way of working. Our goal was to show participants how these technologies can be deployed concretely in organisations and where the advantages over purely passive chatbots like ChatGPT lie. Stefan, head of the AI team at bbv, emphasised at the start of our meeting that we wanted to focus on the interplay between AI agents and operational processes. He invited me as a discussion partner because I myself work closely on such use cases in several AI projects.",[27,2048,2049],{},[11,2050,2051],{},"\"We are giving you an exclusive look into our BBV AI Hub today, starting from the challenges we encounter daily with our clients.\"",[11,2053,2054],{},"This quote from Stefan outlines the core of our event. Our aim was to show what generative AI can practically achieve in an enterprise context. Participants wanted to know how to introduce agentic systems in a way that makes them not merely a nice add-on but a genuine relief in day-to-day operations.",[15,2056],{":video-id":17},[19,2058,2060],{"id":2059},"ai-agents-as-autonomous-problem-solvers","AI Agents as Autonomous Problem Solvers",[11,2062,2063],{},"In the webinar I first explained the term \"AI agent\". It refers to virtual units capable of handling their tasks independently, much like human employees. This means they do not just answer a question but receive a defined goal and independently work out how best to achieve it. Stefan put it this way in the conversation:",[27,2065,2066],{},[11,2067,2068],{},"\"Reacting autonomously to inputs, following a specific workflow and delivering the desired results on the basis of that workflow.\"",[11,2070,2071],{},"Behind this lies a sequence of steps similar to the human problem-solving process. The agent recognises a situation, interprets it and in the next step determines which actions make sense. It then carries out those actions and checks whether they lead to the desired result. If gaps or errors appear, it adjusts its approach. I emphasised that this goes far beyond simple chatbot logic. AI agents work according to an agentic workflow whose core elements are perception, planning, action and feedback loops.",[11,2073,2074],{},"To make this more tangible, I described an example of inventory management at a logistics company. An AI agent continuously checks how much material is still available, forecasts requirements based on historical data and, if necessary, independently initiates a new order. It even decides when which quantities make sense and automatically notifies the responsible department.",[27,2076,2077],{},[11,2078,2079],{},"\"An AI agent could monitor stock levels in real time, automatically trigger actions when levels are low and thereby save time and resources in the long run.\"",[11,2081,2082],{},"This is intended to illustrate that these programmes are capable of considerably more than pure chat functions.",[19,2084,2086],{"id":2085},"the-bbv-ai-hub-technology-platform-and-opportunities","The BBV AI Hub: Technology Platform and Opportunities",[11,2088,2089],{},"In a second step I presented the BBV AI Hub. This is a technology platform we developed to programme new AI agents as quickly as possible and embed them in existing IT landscapes. Anyone asking why such a thing is needed: I explained that classic software architectures are not always designed for the speed at which AI applications evolve.",[11,2091,2092],{},"The AI Hub brings clarity to the management of data and connections to various language models. This means one can opt for commercial services such as OpenAI or Google Cloud, or run open-source models in one's own data centre. For some organisations the latter is indispensable because sensitive data must not leave the company network. A further advantage lies in scalability. When an agent has to process requests very frequently, we can quickly determine whether a more powerful environment or a different model would be more appropriate.",[19,2094,2096],{"id":2095},"challenges-in-client-projects","Challenges in Client Projects",[11,2098,2099],{},"I am often asked where the biggest hurdles lie in developing these AI solutions in day-to-day practice. My answer is always that we as an AI team have the purely technical aspects well under control, but the factual correctness of results must be verified with the support of our clients. In the inventory management example it is clear: we do not know every detail of every sector. This means that an AI agent may work formally correctly but there is always a possibility that a sector-specific rule gets overlooked.",[11,2101,2102],{},"In the webinar we showed how we address this problem. We log the individual steps of our agents. As soon as an incorrect answer appears, one can trace in an agent logbook which texts or data sources were evaluated and whether the system perhaps used outdated information. The nice thing about this is that one learns a great deal from these analyses and continuously improves the agents based on client feedback.",[27,2104,2105],{},[11,2106,2107],{},"\"Particularly with critical decisions we want to ensure that the AI agent tells us which document it is relying on.\"",[11,2109,2110],{},"That is how I explained it in the webinar. When an automated action has far-reaching consequences, transparency is essential.",[11,2112,2113],{},"Collaboration Between Agents\nA topic that interested participants greatly afterwards was the question of whether multiple AI agents can communicate with one another. I confirmed that we have already developed solutions for this.",[11,2115,2116],{},"It is possible for a superordinate coordinator agent to recognise that a task requires several specialist agents. It then forwards requests to the relevant agents and collects their partial results until an overall solution is available. This principle follows the same logic as in real teams, where different experts work together.",[11,2118,2119],{},"In the webinar I showed, for example, how an agent responsible for monitoring inventory levels can cooperate with another agent that initiates orders or tenders. The result is an AI agent team capable of making independent decisions without a human having to control every intermediate step.",[19,2121,2123],{"id":2122},"roles-rights-and-data-access","Roles, Rights and Data Access",[11,2125,2126],{},"In many client conversations the question of access control comes up quickly. Employees should only see information that they are permitted to access according to their role. In such cases I emphasise that we always define agents quite clearly. An agent receives exactly the data sources it needs for its task, no more and no less.",[11,2128,2129],{},"The AI Hub also provides connections to existing access and rights systems. Anyone not permitted to use the agent does not see it and therefore cannot read critical data. I pointed out that this is a key factor for AI systems to integrate securely into operational life in the long run.",[19,2131,2133],{"id":2132},"finding-knowledge-rather-than-searching-for-it","Finding Knowledge Rather Than Searching for It",[11,2135,2136],{},"Many organisations have extensive document or knowledge repositories. According to studies, employees waste valuable time every day because they do not know exactly where a piece of information is stored. In the webinar Stefan mentioned a statistic showing that employees sometimes spend 30 minutes a day just searching for documents. I explained how AI agents in combination with so-called vector-based databases can provide a remedy.",[11,2138,2139],{},"When a question is asked, the agent searches the relevant sources and very quickly finds those text passages that are suited to answering it. These are far more complex search strategies than a simple full-text search. By evaluating context, the semantic proximity between terms, the agent finds the right passages even with synonyms or related topics. What is decisive is that we continuously improve this system and regularly check whether the agent is delivering correct results.",[19,2141,2143],{"id":2142},"live-demo-transparency-and-traceability","Live Demo: Transparency and Traceability",[11,2145,2146],{},"During the webinar I showed in a live demo how an AI agent works in a question-and-answer scenario. Participants could see that the agent goes through several intermediate steps with each question. It checks whether it is responsible for the question at hand, searches for relevant text passages, filters them and only generates an answer at the end.",[11,2148,2149],{},"We also demonstrated that all sub-steps can be viewed in the agent logbook, for example which documents or passages the agent used as a source for its answer. This transparency gives users confidence, because they can trace when needed why an answer was formulated in a particular way.",[19,2151,2153],{"id":2152},"outlook-and-own-use-cases","Outlook and Own Use Cases",[11,2155,2156],{},"At the end of the webinar Stefan asked me which AI agents we use internally ourselves. I told him about a so-called wiki agent we developed that can access our company-specific guidelines and our intranet. Employees looking for specific HR regulations or IT policies no longer have to manually search multiple platforms.",[11,2158,2159],{},"Beyond that, we support teams with the creation of project descriptions by having an agent automatically collect the relevant key data and insert it sensibly into a document. In addition, we have an agent that bundles internal AI knowledge. It answers colleagues' questions before those questions reach us as AI developers.",[27,2161,2162],{},[11,2163,2164],{},"\"We see in practice that many questions concentrate on a small core of company knowledge. That is exactly where it is particularly worthwhile to create a cross-departmental, central point of contact with AI agents.\"",[11,2166,2167],{},"I assured participants that we will continue to expand this principle in future, for example with agents that automate ordering processes or ease the creation of documentation entries in complex projects.",[11,2169,2170],{},"Stefan concluded by pointing to the Swiss AI Impact Report, which we would be publishing shortly. In it we evaluate a study we conducted on AI use cases in Swiss organisations. We want to share our findings in a future webinar and show which experiences Swiss companies have already had with generative AI.",[19,2172,902],{"id":901},[11,2174,2175],{},"I have the impression that the evening was an eye-opener for many participants. We were able to illustrate clearly that generative AI systems are already far more than simple chatbots. AI agents can address concrete operational challenges that reach deep into everyday business. Above all, I emphasised the importance of transparency. Only when it is clear how an agent arrives at its recommendations are organisations willing to hand over responsibility to such autonomous systems.",[11,2177,2178],{},"For me personally it remains clear that AI agents are only as smart as the knowledge made accessible to them. In collaboration with clients, however, we experience how quickly the full potential unfolds when the right information is provided and the right role and security concepts are defined. It is precisely this combination of technical innovation and thoughtful integration into existing operational processes that fascinates me.",{"title":124,"searchDepth":125,"depth":125,"links":2180},[2181,2182,2183,2184,2185,2186,2187,2188],{"id":2059,"depth":128,"text":2060},{"id":2085,"depth":128,"text":2086},{"id":2095,"depth":128,"text":2096},{"id":2122,"depth":128,"text":2123},{"id":2132,"depth":128,"text":2133},{"id":2142,"depth":128,"text":2143},{"id":2152,"depth":128,"text":2153},{"id":901,"depth":128,"text":902},"Generative AI and AI agents are at the heart of an agentic way of working in which virtual systems\nsupport business processes autonomously. The BBV AI Hub technology platform shows how data sources,\nlanguage models and security concepts can be intelligently connected. Examples from logistics and HR\nillustrate how comprehensively and transparently AI agents can be deployed in practice.\n","2024-09-25","bbv KI Webinar - AI Agents in Practice","\u002Fimages\u002Fpresentations\u002Fwebinar5\u002Fki_agenten_praxis.png",{},"\u002Fpresentations\u002Fen\u002Fwebinar_ki_agenten_in_der_praxis",{"title":2041,"description":2189},"presentations\u002Fen\u002Fwebinar_ki_agenten_in_der_praxis","Transparency, Performance and Innovation with the AI Hub","AI Agents in Practice","1ME9gtgb-f4","137","om1BPTdtbJZKWT1ie4VoMTIuHlCQ6yBIc6gfDajFmUU",{"id":2203,"title":2204,"audience":2205,"body":2206,"carouselItems":136,"companyName":136,"date":136,"description":2317,"end_date":136,"eventDate":2318,"eventDuration":2319,"eventName":2320,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":2321,"mainTitle":136,"meta":2322,"navigation":144,"path":2323,"podcastId":136,"role":136,"seo":2324,"series":136,"start_date":136,"stats":136,"stem":2325,"subtitle":2326,"tags":136,"techs":136,"titleHighlight":2327,"videoId":2328,"views":2329,"watchTime":136,"__hash__":2330},"presentations\u002Fpresentations\u002Fen\u002Fwebinar_ki_als_entwicklungspartner.md","Webinar: AI as Development Partner","ca. 50 participants",{"type":8,"value":2207,"toc":2309},[2208,2211,2213,2217,2220,2225,2228,2232,2235,2238,2243,2246,2251,2255,2258,2261,2266,2269,2273,2276,2279,2284,2287,2291,2294,2297,2302,2306],[11,2209,2210],{},"On 23 April 2025 bbv held another episode of its webinar series on artificial intelligence, this time with a focus on software development. I moderate the session and guide participants through practical insights from the BBV AI Hub. Together with my colleagues Thomas Mannhart, Michèle Fundneider, Noah Hermann and Joel Barmettler we examine how AI is used along the development process, from requirements analysis and coding through to testing, reviews and architecture decisions. The goal is a sober, application-oriented look at which methods and working practices prove their worth in daily use and where deliberate context and quality management remains decisive.",[15,2212],{":video-id":17},[19,2214,2216],{"id":2215},"ai-in-the-development-process-practical-experiences-from-the-bbv-ai-hub","AI in the Development Process: Practical Experiences from the BBV AI Hub",[11,2218,2219],{},"AI has arrived in the everyday life of software development. In the BBV AI Hub it is particularly clear how strongly working methods change when language models not only answer individual questions but participate in structuring, formulating, reviewing and understanding development artefacts. The goal here is not a tool comparison but a practical look at the methods and strategies that prove themselves in the development process, from requirements analysis and code development through to testing and review. The content is aimed at technically experienced people who already know software development processes, whether they are just starting with AI or already have experience.",[27,2221,2222],{},[11,2223,2224],{},"\"Our goal today is to provide practical insights into the possibilities and challenges of integrating AI into the development process.\"",[11,2226,2227],{},"In the webinar several roles from the team have their say in order to make different perspectives visible: condensing requirements from workshops, using AI productively for coding, getting up to speed faster in new codebases and concepts, and as an architect maintaining an overview of tools, frameworks and system decisions. One thread runs through all contributions: AI delivers the greatest benefit where context is managed cleanly and results remain verifiable.",[19,2229,2231],{"id":2230},"from-workshop-results-to-requirements-making-unstructured-data-quickly-usable","From Workshop Results to Requirements: Making Unstructured Data Quickly Usable",[11,2233,2234],{},"Workshops often produce many artefacts: notes, sketches, post-its, flip charts, mockups. The bottleneck lies less in a lack of information than in its preparation: what was decided, how can it be played back in a structured way to the team and client, and how do actionable requirements emerge from it? Thomas describes a workflow that provides significant relief here.",[11,2236,2237],{},"One important step is capturing content directly after the workshop via speech-to-text. Instead of later laboriously reconstructing from memory, a rough draft is created early, deliberately unstructured but complete. This draft is then brought into a structured form by assistants, for example as epics and requirements including acceptance criteria. Additionally, photos of post-its or flip charts can serve as context: models extract content from images, cluster it and deliver summaries that in turn can be used for requirements, project logs or documentation.",[27,2239,2240],{},[11,2241,2242],{},"\"I just dictate everything that comes to mind about this workshop, and so I have it stored in an unstructured form for the first time.\"",[11,2244,2245],{},"A realistic view is part of reliability: for smaller, clearly bounded features the automatic creation works particularly well; with larger scope, the need for post-editing increases. A practical bonus is that AI does not only \"generate\" but also helps with reviewing: suggestions for non-functional requirements, notes on gaps, contradictions or overly vague formulations.",[27,2247,2248],{},[11,2249,2250],{},"\"Hey, do you notice anything else? Is something contradictory? Is something too vague?\"",[19,2252,2254],{"id":2253},"ai-in-coding-choosing-context-deliberately-rather-than-throwing-everything-in","AI in Coding: Choosing Context Deliberately Rather Than \"Throwing Everything In\"",[11,2256,2257],{},"Michelle uses AI broadly in engineering daily life: generating code, refactoring, writing tests, validating ideas, both through IDE-proximate tools (for example Copilot) and through chat interfaces. The biggest challenge is less the model itself than context management. Two extremes recur constantly: too much context (unspecific, unclear where statements come from) and too little context (generic suggestions that do not match guidelines, language or architecture).",[11,2259,2260],{},"The workable solution is deliberate, often manual context management: selecting relevant code sections specifically, formulating goals and boundary conditions clearly, providing examples and consistently starting fresh chats when topics change. It is also worthwhile bundling recurring project information in specialised assistants (guidelines, output formats, typical constraints) to avoid starting from scratch with every task.",[27,2262,2263],{},[11,2264,2265],{},"\"Clarity over data flood: better less, but cleanly explained context.\"",[11,2267,2268],{},"Web search in AI tools is also used selectively: disabled by default to keep context controllable, and only enabled when currency is genuinely needed (for example library bugs, new releases). This follows the same principle: open context deliberately, not expand it randomly.",[19,2270,2272],{"id":2271},"onboarding-learning-and-reviews-ai-as-navigator-and-pre-check","Onboarding, Learning and Reviews: AI as Navigator and Pre-Check",[11,2274,2275],{},"From a junior perspective (Noah), AI is particularly helpful when getting to grips with large codebases. When you do not yet know where something is implemented, AI can help map abstract questions (\"how do we handle persistence?\") to concrete places in the code. This creates orientation faster without constantly blocking the team.",[11,2277,2278],{},"For new concepts not directly present in the codebase, a second technique helps: telling the AI how you want to learn. Noah uses the metaphor of a knowledge tree (fundamentals, core concepts, details) and controls the pace explicitly: only move on when the current concept is truly understood. The result is a dialogic learning process rather than an overloaded \"everything at once\" text.",[27,2280,2281],{},[11,2282,2283],{},"\"Only move on to the next concept when I really say I have understood it.\"",[11,2285,2286],{},"AI is also used as a pre-review: pull requests can be automatically summarised and checked for anomalies (for example potential security issues). This does not replace a human review but helps with triage and preparation, for instance to see early how large a PR is and where the risky parts lie.",[19,2288,2290],{"id":2289},"architecture-and-tool-landscape-from-the-big-picture-to-implementation","Architecture and Tool Landscape: From the Big Picture to Implementation",[11,2292,2293],{},"As an AI architect, Joel works heavily at the start of client projects: requirements, infrastructure, data protection, model selection and the question of which problems should actually be solved with AI. At the same time the platform perspective must consider which tools and standards will be \"relevant in six months\". To handle information overflows, deep research workflows help. What is decisive here is source control. Rather than broad web searches, trustworthy sources are preferred (papers, GitHub, selected communities) to reduce the risk of misinformation.",[11,2295,2296],{},"After the pre-selection comes the deep dive: new open-source projects are often poorly documented, so AI is particularly helpful for navigating codebases and finding relevant implementation points. For system questions a very large context window can be useful (for example a condensed codebase to identify sensible integration points). For the actual coding, context is then reduced again so that precise changes remain possible. The switch between \"lots of context for overview\" and \"little context for implementation\" is a deliberate pattern here.",[27,2298,2299],{},[11,2300,2301],{},"\"At every stage I use different tools, from the overview to the specific implementation.\"",[19,2303,2305],{"id":2304},"models-benchmarks-and-data-protection-pragmatic-guardrails","Models, Benchmarks and Data Protection: Pragmatic Guardrails",[11,2307,2308],{},"When comparing models, benchmarks and leaderboards help, but they are not infallible: models can be optimised for benchmarks. Blind tests with user evaluations are more robust but less specific. In practice a mixture works well: rough classification via leaderboards, then cross-checking with coding or reasoning benchmarks, and finally a personal comparison with typical everyday tasks (the same question to multiple models). Data protection remains a framework that co-determines tool selection and hosting options, especially in client projects.",{"title":124,"searchDepth":125,"depth":125,"links":2310},[2311,2312,2313,2314,2315,2316],{"id":2215,"depth":128,"text":2216},{"id":2230,"depth":128,"text":2231},{"id":2253,"depth":128,"text":2254},{"id":2271,"depth":128,"text":2272},{"id":2289,"depth":128,"text":2290},{"id":2304,"depth":128,"text":2305},"The webinar focuses on practical uses of AI in software development, from requirements analysis through to\ntesting and code reviews. It shows how workshop artefacts such as notes, post-its and sketches can be\nstructured with speech-to-text and multimodal models and converted into requirements. In coding, the focus\nis on deliberate context management, sensible prompting strategies and the use of specialised assistants.\nFor onboarding and knowledge building, AI is used as a navigation and learning aid in large codebases and\nas a step-by-step explanation partner. From an architecture perspective, the session covers tool research\nwith controlled sources, systematic exploration of new codebases and switching between overview and\nimplementation context.\n","2025-04-23","43 Min.","bbv KI Webinar - AI as Development Partner","\u002Fimages\u002Fpresentations\u002Fwebinar7\u002Fai_in_software_1.png",{},"\u002Fpresentations\u002Fen\u002Fwebinar_ki_als_entwicklungspartner",{"title":2204,"description":2317},"presentations\u002Fen\u002Fwebinar_ki_als_entwicklungspartner","Tools, Techniques and Team Integration","AI as Development Partner","CCc3SREchJE","292","CP34YlNzLm3cwAKkYNIbIU--Hw6lRF4rC4X_shSrq74",{"id":2332,"title":2333,"audience":2334,"body":2335,"carouselItems":136,"companyName":136,"date":136,"description":2317,"end_date":136,"eventDate":2500,"eventDuration":2501,"eventName":2502,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":2503,"mainTitle":136,"meta":2504,"navigation":144,"path":2505,"podcastId":136,"role":136,"seo":2506,"series":136,"start_date":136,"stats":136,"stem":2507,"subtitle":2508,"tags":136,"techs":136,"titleHighlight":2509,"videoId":2510,"views":2511,"watchTime":136,"__hash__":2512},"presentations\u002Fpresentations\u002Fen\u002Fwebinar_ki_use_cases.md","Webinar: Practical AI","ca. 65 participants",{"type":8,"value":2336,"toc":2489},[2337,2340,2342,2346,2349,2354,2357,2361,2364,2367,2372,2375,2379,2382,2385,2388,2391,2396,2399,2403,2406,2409,2412,2415,2419,2422,2427,2430,2434,2437,2440,2443,2447,2450,2453,2458,2461,2465,2468,2471,2476,2479,2483,2486],[11,2338,2339],{},"On 18 March 2026 bbv held another episode of its webinar series on artificial intelligence. This time the focus is on how concrete AI use cases are built in organisations and what role a technology platform plays in that. Together with Emre I put various application scenarios from the Swiss AI Hub context into perspective and show how assistants and agents can be integrated into existing processes. The goal is not a look at individual tools or demos but a sober, practical assessment of how AI systems can become usable in real enterprise contexts.",[15,2341],{":video-id":17},[19,2343,2345],{"id":2344},"ai-use-cases-in-the-enterprise-platforms-processes-and-concrete-applications","AI Use Cases in the Enterprise: Platforms, Processes and Concrete Applications",[11,2347,2348],{},"At the centre of the webinar for me is a simple guiding idea: artificial intelligence unfolds its value in an enterprise context not where as much as possible is said about visions, but where concrete processes, clean technical integration and reliable use cases come together. It is precisely from this perspective that I present several examples from the Swiss AI Hub environment. The aim is not a product showcase or spectacular demos but the question of how AI applications can fit into existing enterprise realities. What is decisive is less the language model alone and more the ability to connect data sources, systems, permissions, processes and human expertise in a meaningful way.",[27,2350,2351],{},[11,2352,2353],{},"\"It is all about consolidated knowledge.\"",[11,2355,2356],{},"At the start we frame the market using a simple model. At the very bottom sit the foundation models, the actual language models such as GPT, Gemini or open-source models. Above them are interaction elements such as chat interfaces, web search or image generation, the tools with which people work with these models. Above that sit integrated enterprise solutions such as Microsoft Copilot or Gemini in workspace applications. Still more concrete are specialised AI apps that solve individual problems in clearly defined domains. For many organisations this is sufficient as long as their requirements can be covered by existing solutions. Things become more difficult where processes are highly individual, historically grown or strongly company-specific. In that case, introducing a ready-made product often falls short. A proprietary technological foundation is then needed on which customer-specific applications can be built.",[19,2358,2360],{"id":2359},"an-ai-platform-as-the-foundation-for-individual-business-processes","An AI Platform as the Foundation for Individual Business Processes",[11,2362,2363],{},"This is precisely where a technology platform comes in. Such a platform does not simply provide access to a single model; it creates the prerequisites for using different models appropriately for each task, integrating data in a privacy-compliant way and making AI capabilities available where employees already work. This turns AI not into an isolated add-on tool but into an integrated component of existing workflows.",[11,2365,2366],{},"An important distinction in the webinar is that between assistants and agents. Both are based on language models and can be equipped with tools, data or permissions. The difference lies for me above all in their role within a process. Assistants are more passive. They are triggered by people or other units and then complete a clearly formulated task. Agents, by contrast, are more deeply embedded in processes and can work more autonomously. They do not only become active on request but at defined points in a business process.",[27,2368,2369],{},[11,2370,2371],{},"\"An assistant works in support of the employee. An agent slots into a business process.\"",[11,2373,2374],{},"This distinction is not meant purely technically but helps to describe AI systems clearly in business terms. An assistant supports employees directly in their work. An agent becomes part of a defined workflow that can also function without constant human input. This difference becomes visible again and again in the examples presented.",[19,2376,2378],{"id":2377},"bmd-supporting-service-technicians-with-context-based-knowledge","BMD: Supporting Service Technicians with Context-Based Knowledge",[11,2380,2381],{},"A particularly illustrative use case comes from the area of crane systems and lifting solutions. The challenge there is that service staff in the field need to access very large, heterogeneous knowledge quickly when faults occur. In addition to proprietary systems, products from third-party suppliers are also used, generating extensive documentation, circuit diagrams, manuals and service-related information. When a problem arises on site, the back office is frequently contacted. Experienced key people sit there with extensive knowledge, but they too must search through large data stocks before they can help. In a situation where a system has stopped, exactly this delay is particularly costly.",[11,2383,2384],{},"The first stage of the solution therefore consists of an AI assistant based on a classic RAG concept. A language model is combined with a company-specific knowledge pool. The system thereby delivers not only general answers but context-based information from exactly those documents and sources relevant to the specific customer and the specific system. The real value, however, only arises through the additional connection to history and customer data. A good assistant does not just name a theoretically appropriate measure; it can also proactively provide hints about known past faults or specific weaknesses in a particular system.",[11,2386,2387],{},"What is decisive is that this quality does not arise through the language model alone. It rests on the technical connection of knowledge data, customer data, ticketing system and documentation. Only through this integration does a general AI answer become reliable support for the specific service case.",[11,2389,2390],{},"In a second stage, the system is extended by an agent. When the assistant cannot find a sufficiently reliable answer or a defined confidence value is not reached, the agent brings in a human expert. This knowledge flows back not only into the current answer but is simultaneously transferred to the central knowledge pool. Expert knowledge is thereby made systematically available step by step rather than remaining locked in individual minds.",[27,2392,2393],{},[11,2394,2395],{},"\"The knowledge does not stay tied to the person.\"",[11,2397,2398],{},"This point is central, especially in times of skills shortages and retirements. AI here is understood not as a replacement for expertise but as a mechanism for retaining expertise within the organisation, making it accessible and usable in the long term.",[19,2400,2402],{"id":2401},"laboratory-automation-bringing-together-support-knowledge-from-multiple-sources","Laboratory Automation: Bringing Together Support Knowledge from Multiple Sources",[11,2404,2405],{},"A further case shows the same basic idea in a different domain, namely in the technical support of an internationally operating company in the area of laboratory automation. The starting position is particularly demanding here because many large customers are served worldwide and almost every system is unique. Relevant knowledge is distributed across systems such as Jira, SharePoint and Confluence, partly in different languages. When a problem arises, a time-consuming search across multiple sources frequently begins. This is particularly burdensome and inefficient for support staff.",[11,2407,2408],{},"Here the solution consists of embedding the AI directly in the existing support process. Instead of opening a separate tool, the support is immediately accessible in Jira or alternatively via Teams. As soon as a ticket is opened, a multi-source search starts in the background. Relevant content from different systems is consolidated, normalised linguistically and provided in the form of a solution proposal. The AI thereby works exactly where the staff are already working.",[11,2410,2411],{},"Such an approach is helpful not only because it saves time. It also increases acceptance because the support takes place in familiar systems. AI is not introduced as a new target system but as an extension of existing work steps. This integration capability is in practice often more decisive than raw model performance.",[11,2413,2414],{},"The concept can also be extended toward the customer. In a customer portal, an AI system can query relevant information before a ticket is even created, collect error reports and deliver initial solution approaches. If the matter cannot be resolved in this way, a ticket with a much better information basis is created. This also speeds up support considerably because follow-up questions are reduced and important information is structured and captured earlier.",[19,2416,2418],{"id":2417},"data-access-rights-and-separation-of-customer-knowledge","Data Access, Rights and Separation of Customer Knowledge",[11,2420,2421],{},"This case in particular shows very clearly how important governance, access control and clean data management are. When multiple customers, sensitive information and global organisations are involved, a system must never accidentally use one customer's knowledge in another context. Such requirements cannot be solved by good prompting alone. They require precise technical architecture, clean rights management and clearly defined data pipelines.",[27,2423,2424],{},[11,2425,2426],{},"\"You can cause very many data protection or compliance problems if you do not use these data pipelines correctly.\"",[11,2428,2429],{},"For me this is one of the most important points of the entire webinar. In many discussions about generative AI, the model is in the foreground. In real enterprise projects, however, it is often the surrounding architecture that is the decisive success factor. It is that architecture which ensures that answers are relevant, traceable and organisationally permissible in the first place.",[19,2431,2433],{"id":2432},"fmh-making-complex-billing-rules-for-medical-services-accessible","FMH: Making Complex Billing Rules for Medical Services Accessible",[11,2435,2436],{},"A third use case concerns the medical domain and the question of how complex billing rules for medical services can be made accessible. Doctors describe concrete consultations, must bill them correctly and consult extensive rulebooks and manuals to do so. When ambiguities arise, queries frequently land with specialist staff who must answer them manually. This costs time on both sides and ties up valuable expertise in repetitive clarifications.",[11,2438,2439],{},"The solution sketched for this works with multiple AI units. An assistant receives the input, structures it and breaks complex descriptions into sub-questions. The sub-questions are then forwarded to specialised agents that each access different knowledge areas, for example manuals or tariff rules. The results are then reassembled into a comprehensible answer.",[11,2441,2442],{},"Here too the aspiration is not to create a black box but to provide traceable support with source references. Transparency is central, especially in sensitive specialist domains. Users must be able to see where a statement comes from and on what basis it was generated. When certain answers cannot be generated with sufficient certainty, the system should not simply continue speculating but escalate the case to human specialists. AI thereby remains a tool within a process for which specialists bear responsibility.",[19,2444,2446],{"id":2445},"not-every-sensible-use-case-is-a-good-generative-ai-use-case","Not Every Sensible Use Case Is a Good Generative AI Use Case",[11,2448,2449],{},"A particularly important point of the webinar for me is to consciously correct false expectations. Not every sensible digital use case is automatically a good generative AI use case. This is made clear through a deliberately constructed counter-example. Tasks that at their core rely on numerical calculations, statistical models or deterministic forecasts can perhaps be combined with AI components, but generative language models are not automatically the right tool for them.",[11,2451,2452],{},"A language model can understand text, structure it and work with context. But it is not a substitute for precise mathematical models, classical statistics or robust calculation pipelines. In organisations the notion quickly arises that one can give a powerful model all the PDFs, Excels and rules, and it will then solve all problems. This expectation is unrealistic.",[27,2454,2455],{},[11,2456,2457],{},"\"LLMs are language models, not calculators.\"",[11,2459,2460],{},"Linked to this is another widespread misconception: the assumption that a single central chat window is sufficient for AI to correctly understand and process all kinds of requests in an organisation. This too falls short. Whether a password is reset, a ticket classified, a document summarised or a specific business process triggered must be cleanly connected and controlled technically. Language models alone do not orchestrate secure business processes. Only interfaces, permissions, defined workflows and complementary software logic turn this into a reliable application.",[19,2462,2464],{"id":2463},"requirements-test-cases-and-feedback-loops-rather-than-blind-go-live","Requirements, Test Cases and Feedback Loops Rather Than Blind Go-Live",[11,2466,2467],{},"A further focus is on the working method around AI projects themselves. A system does not simply go live and then work stably as desired. Good AI applications are built iteratively. They need requirements, acceptance criteria, test cases, responsible people and feedback loops. Particularly important is the question of what actually constitutes a good answer. Correctness alone is not enough. Style, completeness, brevity, source references or proactive hints also belong to the quality characteristics that must be clarified in advance.",[11,2469,2470],{},"Without this groundwork, even a technically sound system risks failing through lack of acceptance. When users expect perfect answers from the start and receive no realistic framing, the system is quickly perceived as unusable. That is why expectation management, monitoring and the continuous improvement of data structures and processes are so important.",[27,2472,2473],{},[11,2474,2475],{},"\"Day 1 is not the end of the project.\"",[11,2477,2478],{},"Added to this is the fact that existing legacy systems can often not be made usable by AI without further effort. Historically grown interfaces, proprietary databases or poorly structured legacy systems must first be translated, opened up or technically wrapped before an agent or assistant can work with them meaningfully. This too is not a side issue but a large part of the actual project work.",[19,2480,2482],{"id":2481},"ai-in-the-enterprise-needs-more-than-a-strong-model","AI in the Enterprise Needs More Than a Strong Model",[11,2484,2485],{},"In the end, from all this a clear picture emerges for me: valuable AI applications in enterprises do not arise from the largest possible promises but from precisely tailored solutions. What is decisive is which problem actually needs to be solved, which data are available for that, which systems must be connected and how human expertise can be meaningfully supplemented. AI is strong when it is integrated into processes, understands context, makes knowledge accessible and relieves employees in targeted ways.",[11,2487,2488],{},"The webinar therefore shows less individual tools and more a basic pattern that runs through multiple use cases: AI needs a clean technical foundation, clear roles between assistants and agents, controlled data access and a realistic understanding of its limits. It is precisely in this interplay that the difference between hype and productive application lies.",{"title":124,"searchDepth":125,"depth":125,"links":2490},[2491,2492,2493,2494,2495,2496,2497,2498,2499],{"id":2344,"depth":128,"text":2345},{"id":2359,"depth":128,"text":2360},{"id":2377,"depth":128,"text":2378},{"id":2401,"depth":128,"text":2402},{"id":2417,"depth":128,"text":2418},{"id":2432,"depth":128,"text":2433},{"id":2445,"depth":128,"text":2446},{"id":2463,"depth":128,"text":2464},{"id":2481,"depth":128,"text":2482},"2026-03-18","59 Min.","bbv KI Webinar - Practical AI","\u002Fimages\u002Fpresentations\u002Fwebinar10\u002Fai_use_cases_1.png",{},"\u002Fpresentations\u002Fen\u002Fwebinar_ki_use_cases",{"title":2333,"description":2317},"presentations\u002Fen\u002Fwebinar_ki_use_cases","Pioneering Use Cases with the Swiss AI Hub","Practical AI","EGSTDTqYHjg","8373","FBKQ8uGhLkUv2Vkj9a8zO85I0-7bmM8tnvnCAB3Bh3Y",{"id":2514,"title":2515,"audience":2516,"body":2517,"carouselItems":136,"companyName":136,"date":136,"description":2604,"end_date":136,"eventDate":2605,"eventDuration":2606,"eventName":2607,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":2608,"mainTitle":136,"meta":2609,"navigation":144,"path":2610,"podcastId":136,"role":136,"seo":2611,"series":136,"start_date":136,"stats":136,"stem":2612,"subtitle":2613,"tags":136,"techs":136,"titleHighlight":2614,"videoId":2615,"views":2616,"watchTime":136,"__hash__":2617},"presentations\u002Fpresentations\u002Fen\u002Fwebinar_schwarmintelligenz.md","Webinar: Swarm Intelligence","ca. 60 participants",{"type":8,"value":2518,"toc":2602},[2519,2522,2524,2529,2532,2535,2538,2543,2546,2549,2554,2557,2560,2563,2566,2569,2572,2577,2580,2583,2588,2591,2594,2599],[11,2520,2521],{},"In our webinar on \"Swarm Intelligence: How AI Agents Transform Your Organisation\" I explored together with Joel Barmettler how organisations can concretely benefit from generative AI and why now is the right moment to get started. Our aim was not only to question the current AI hype but above all to show that AI has long since become an effective tool in everyday business operations. Many organisations have already recognised that artificial intelligence can be used to automate processes and open up new business areas.",[15,2523],{":video-id":17},[27,2525,2526],{},[11,2527,2528],{},"\"AI agents bring real value to organisations when you define tasks clearly and give them the right tools.\"",[11,2530,2531],{},"Through this focus on the practical benefit of AI we created a framework in which questions of feasibility, security and cost control become clear. Especially in the current phase, in which generative AI is advancing at a rapid pace, it makes sense to plan the entry deliberately and link it to a clear strategy.",[11,2533,2534],{},"The Path to Generative AI\nIn the next step we looked at the recent research history to show how language models have developed from GPT-2 through GPT-3 (which many know through ChatGPT) to the multi-agent systems currently being discussed. This development is remarkable because it makes clear why AI agents can achieve far more today than just a few years ago.",[11,2536,2537],{},"The concept of few-shot learning was presented, for instance. This concerns how a large language model like GPT-3 can learn new tasks without being completely retrained each time. It is sufficient to give the model a few examples for it to recognise patterns and transfer them independently. Even more interesting are the so-called ReAct approaches, in which language models are not limited to pure text comprehension but can additionally \"think\" and \"tap into\" external tools or data sources to handle complex tasks.",[27,2539,2540],{},[11,2541,2542],{},"\"The ability of AI models to independently 'think things through' and act appropriately opens up entirely new possibilities when it comes to automating high-level tasks.\"",[11,2544,2545],{},"This means we are able to implement not only simple question-and-answer scenarios but to create agents that independently plan intermediate steps. They fetch the necessary information themselves, for example from internal company documents, or access external services to complete their task.",[11,2547,2548],{},"From Individual Agents to the AI Swarm\nAt this point we made the connection to the idea of deploying AI agents not in isolation but networked with one another. This creates a form of swarm intelligence where each agent is responsible for a clearly defined task while simultaneously having the ability to cooperate with other agents or coordinate with them when needed. The result is that an organisation's performance can be effectively multiplied because several specialised agents work in parallel or in coordination.",[27,2550,2551],{},[11,2552,2553],{},"\"A single agent can already be helpful, but the real leverage comes when multiple agents act in teams, mutually controlling and complementing each other.\"",[11,2555,2556],{},"We particularly highlighted two basic interaction models. In the first model, the group chat, all agents (and human team members when needed) exchange directly with one another. This resembles the chat groups we already use in everyday working life in Slack or Teams. In the second model there is a coordinator agent that controls all communication and reports results back in a concise, easily understandable form. Which variant is better suited depends on factors such as team culture, working style and specific project requirements.",[11,2558,2559],{},"Implementation and Practice\nOf course we also wanted to show how an organisation can harness this potential. In the webinar we presented four consecutive phases that have proven themselves in practice for successfully implementing AI agents step by step.",[11,2561,2562],{},"In the introduction phase, the first step is to assign initial AI agents to relatively narrow, well-definable tasks. This allows both employees and the AI to learn what works and where potential limits lie. This approach builds mutual trust and allows risks to be kept within a controlled framework.",[11,2564,2565],{},"Once initial successes have been achieved, the second phase can expand the autonomy of the agents. The focus here is above all on agents independently recognising knowledge gaps and actively asking experts for new information when something is missing. This process requires some patience from the workforce at times, but sustainably strengthens the agents' self-understanding in day-to-day business.",[11,2567,2568],{},"In the third phase, targeted networking takes place. Here several agents are configured so that they can work on joint projects. For the organisation this means a noticeable efficiency boost, as the agents steer each other to a degree in order to bring in their respective areas of expertise.",[11,2570,2571],{},"Once this collaboration is well established, the next step can be taken in the fourth phase and a superordinate coordinator agent installed. This accelerates complex projects in particular, because no longer does every team member have to understand and select every agent. Instead there is a central point of contact that internally coordinates all required resources, both human and machine.",[27,2573,2574],{},[11,2575,2576],{},"\"The great difference between a single agent and a swarm lies in the decentralised self-organisation and in the interplay of specialised AI systems.\"",[11,2578,2579],{},"As already mentioned, success in this implementation phase depends heavily on security and data protection aspects. We therefore also presented concrete approaches for controlling permissions, ensuring cost transparency and continuously monitoring the quality of agents.",[11,2581,2582],{},"Outlook\nIn the final part of the webinar we envisioned how this interplay can be thought of on an even larger scale. In the near future, companies will not only have their internal AI agents but will simultaneously be able to cooperate with external agents from other organisations. This creates a true AI ecosystem in which specialised agents are offered \"as a service\". The idea that entire companies base themselves on developing and providing AI agents for others is no longer futuristic: the open-source community is already working feverishly on building appropriate multi-agent frameworks.",[27,2584,2585],{},[11,2586,2587],{},"\"We will soon see platforms on which companies can buy or rent AI agents from different providers to get tailored solutions for their individual challenges.\"",[11,2589,2590],{},"The subsequent question round showed that the demand for information in the areas of data protection, on-premise solutions, rights management and cost optimisation is very high. Our conclusion is that it is essential for organisations to define clear processes and a technology stack from the outset, so that the use of AI agents is not only secure but also economically viable.",[11,2592,2593],{},"We look forward to exploring these topics further in the next webinar in our series, particularly the aspect of shared knowledge management between humans and machines. The potential to capture, structure and make company knowledge usable more efficiently is greater than ever with AI agents.",[27,2595,2596],{},[11,2597,2598],{},"\"Those who start now have the chance to actively shape the design of these new value chains and secure a decisive competitive advantage in the long run.\"",[11,2600,2601],{},"We are convinced that AI agents can become a \"game changer\" when they are anchored strategically in an organisation and trust, transparency and appropriate processes are considered in good time.",{"title":124,"searchDepth":125,"depth":125,"links":2603},[],"AI agents are revolutionising work processes and increasing team efficiency. OpenAI impressively demonstrated\nwith Custom GPTs what extraordinary synergies emerge when several of these intelligent assistants work\ntogether. The potential for transforming your organisation is enormous. We explain how generative AI can\nchange your organisation for the better and what steps are needed to deploy AI agents in your organisation\nsuccessfully. We show practically, using concrete examples, how a multi-agent system can be actively used,\nand we also venture a look into the future and what it holds for AI agents.\n","2023-12-13","54 Min.","bbv KI Webinar - Swarm Intelligence","\u002Fimages\u002Fpresentations\u002Fwebinar1\u002Fschwarmintelligenz.png",{},"\u002Fpresentations\u002Fen\u002Fwebinar_schwarmintelligenz",{"title":2515,"description":2604},"presentations\u002Fen\u002Fwebinar_schwarmintelligenz","How AI Agents Transform Your Organisation","Swarm Intelligence","7dEHUcZze9I","875","2dS76AIFN9jNAt-Exi8Vk5BT6BWYEhTD9B1diWyPXBA",{"id":4,"title":5,"audience":6,"body":2619,"carouselItems":136,"companyName":136,"date":136,"description":137,"end_date":136,"eventDate":138,"eventDuration":139,"eventName":140,"extension":141,"highlightWord":136,"icon":136,"images":136,"label":136,"link":136,"logo":142,"mainTitle":136,"meta":2701,"navigation":144,"path":145,"podcastId":136,"role":136,"seo":2702,"series":136,"start_date":136,"stats":136,"stem":147,"subtitle":148,"tags":136,"techs":136,"titleHighlight":149,"videoId":150,"views":151,"watchTime":136,"__hash__":152},{"type":8,"value":2620,"toc":2691},[2621,2623,2625,2627,2629,2633,2635,2637,2639,2641,2643,2645,2649,2651,2653,2655,2657,2659,2661,2663,2665,2669,2671,2673,2675,2677,2679,2683,2685,2687,2689],[11,2622,13],{},[15,2624],{":video-id":17},[19,2626,22],{"id":21},[11,2628,25],{},[27,2630,2631],{},[11,2632,31],{},[11,2634,34],{},[19,2636,38],{"id":37},[11,2638,41],{},[11,2640,44],{},[19,2642,48],{"id":47},[11,2644,51],{},[27,2646,2647],{},[11,2648,56],{},[11,2650,59],{},[19,2652,63],{"id":62},[11,2654,66],{},[11,2656,69],{},[19,2658,73],{"id":72},[11,2660,76],{},[11,2662,79],{},[11,2664,82],{},[27,2666,2667],{},[11,2668,87],{},[19,2670,91],{"id":90},[11,2672,94],{},[19,2674,98],{"id":97},[11,2676,101],{},[11,2678,104],{},[27,2680,2681],{},[11,2682,109],{},[11,2684,112],{},[19,2686,116],{"id":115},[11,2688,119],{},[11,2690,122],{},{"title":124,"searchDepth":125,"depth":125,"links":2692},[2693,2694,2695,2696,2697,2698,2699,2700],{"id":21,"depth":128,"text":22},{"id":37,"depth":128,"text":38},{"id":47,"depth":128,"text":48},{"id":62,"depth":128,"text":63},{"id":72,"depth":128,"text":73},{"id":90,"depth":128,"text":91},{"id":97,"depth":128,"text":98},{"id":115,"depth":128,"text":116},{},{"title":5,"description":137},1779977583404]