Greta Buschhaus
Reporter / Moderator
Sessions (3)
Sessions featuring Greta Buschhaus at Bits & Pretzels 2026 | Founders Festival in Munich
Designing the AI-Native Organisation
The core message of the panel was that becoming AI-native is not about adding individual chatbots to existing work, but about redesigning organisational workflows, governance, collaboration, and accountability around AI. Greta Buschhaus framed the discussion by noting that many organisations have run AI pilots, but fewer have changed how the organisation actually works. The panel then moved from the broad claim that AI is a “10x efficiency booster” to the practical question of what changes once companies move beyond pilots and PowerPoint: which workflows shift, which roles evolve, and where human accountability must remain. Andrew McCarthy described AI adoption as progressing through four levels. The first is a single-player chatbot; the second is an assistant for tasks such as enterprise search or drafting reports; the third treats AI as a teammate that joins meetings, captures notes, drafts summaries, and starts follow-up tasks; and the fourth is a full end-to-end system, such as an IT help desk built on Notion. He argued that many organisations struggle to move from level two to level three because that transition shifts AI from individual productivity to team collaboration, where broader organisational value is captured. He also compared today’s AI interfaces to early television, arguing that many tools still resemble chatbots because the first dominant interface shaped expectations. Stefanie Kemp grounded the discussion in healthcare, where the pressure is not primarily competitiveness but a projected shortage of 500,000 people in care and physician roles by the end of 2030. For Sana Kliniken, AI is needed to reduce effortful administrative work, including language capture, speech, settlement, accounting systems, and patient information through portals. She cautioned that AI cannot be adopted as isolated vertical solutions alone: organisations must think early about governance, what data trains the AI, who has transparency into processes, and how those processes will run at scale in a complex organisation of 43,000 people. Timo Uge emphasised adoption as a leadership and mindset challenge. Leaders need to use AI visibly, and organisations must be honest that some service roles will change as tasks that once took two weeks can be done in hours. He urged people to reskill seriously and avoid scattered proofs of concept without orchestration. The discussion ended on hard limits: Stefanie insisted that in healthcare, “someone needs to click the last key” because physicians must accept AI recommendations. Andrew focused on automating knowledge-work meeting processes, while Timo argued that regulation, the EU AI Act, and legal responsibility currently define limits, even if future AI systems may sometimes make better decisions than exhausted humans.
Industrial AI: From Experimentation to Execution
The core message of the session was that enterprise AI value will not come simply from using the same public models as everyone else; it will come from applying AI to proprietary business data, process knowledge, industry context, and governance. Greta Buschhaus introduced Thomas Saueressig, Chief Customer Officer and Executive Board member at SAP, as someone who has sat on both sides of the CIO conversation. Saueressig opened with a personal example: he uses AI agents each morning to scan the web for news about SAP and himself, summarize coverage, analyze sentiment, and correlate weekly sentiment with the stock price. His point was practical rather than futuristic: agents already remove repetitive information-gathering work and create more immediate insight.
AI Sovereignty in Practice: What Model Independence Actually Requires
The core message of this conversation was that AI sovereignty is not achieved by a slogan or a legal label alone; in Langdock’s case, it required restructuring ownership, reducing jurisdictional ambiguity, and building toward greater control over the model and inference stack. Lennard Schmidt explained that Langdock, founded three years earlier and initially structured with a US Delaware C Corp because of its Y Combinator acceptance, recently completed a nine-month process to flip into a European SE. He said the move cost an eight-digit amount of money, but it created what he sees as a more appropriate foundation for a company built in Germany, run by German founders, serving European organizations, and responding to growing customer and market demand for genuinely European AI infrastructure. The legal discussion centered on why a US holding company over a German GmbH created an uncomfortable, if not necessarily clear-cut, sovereignty issue. Schmidt said that in a Patriot Act or Cloud Act scenario, the founders could have been caught between obligations at the US holding level and German obligations at the GmbH level. He emphasized that the GmbH would not legally have been allowed to hand over data and that major customers had accepted the old structure after review. Still, the restructuring was presented as a strategic and emotional choice: it aligned Langdock’s entity structure with its European values, customer expectations, and long-term plans. The conversation then moved from legal sovereignty to technical sovereignty. Schmidt acknowledged that the best frontier models and major hyperscalers remain largely American, naming AWS, GCP, and Azure as important providers with European offerings and data centers. However, he argued that the model layer is becoming more commoditized as open-weight models, particularly Chinese open-weight models, become good enough for many knowledge-work use cases at much lower cost. This shift has encouraged Langdock to use demand from its 300,000 users to build out its own inference capacity, host open-weight models, and give customers more choice over providers and geographies. In the rollout discussion, Schmidt argued that successful enterprise AI adoption is less about a top-down “AI project” and more about people-driven change. His practical pattern starts with a broad, compliant chat application that employees already want because they are using consumer tools such as ChatGPT or Claude outside approved environments. Once internal adoption creates momentum, organizations can connect systems like SharePoint securely, then move toward agentic workflows and process automation with domain experts. He also challenged the idea that German companies should primarily fear AI-driven job losses, saying jobs will transform and that the bigger risk is inertia, especially given Germany’s aging workforce and administrative capacity challenges.