Enterprise AI adoption is stuck - not because models aren't capable enough, but because organizations are solving the wrong problem. This talk argues that the real barriers are structural: a Context Gap (enterprise knowledge is organized for humans, not machines) and an Agency Gap (the role of human judgment in AI-augmented systems remains undefined).
LLMs and frontier AI systems are built on human language - and inherit the full spectrum of human cognition, including its failure modes. AI confabulates the way humans do. Agents game safety guardrails under metric pressure like employees game KPIs. Multi-agent systems drift toward consensus over truth, mirroring organizational groupthink. These aren't bugs - they're structural properties of systems built on human cognitive patterns.
Drawing on cognitive science and real-world enterprise AI deployments at SAP, this talk offers a framework for closing both gaps - and an honest look at what language models alone cannot solve.
