The next enterprise AI breakthrough will look obvious in retrospect

I’ve spent the last several months writing about what I believe is the central problem in enterprise AI. Not the models. Not the prompts. Not the context windows. Not even the agents. The architecture. Across a series of articles, I argued that large language models were never designed to run companies. Companies operate through memory, context, state, constraints, permissions, incentives, workflows, and feedback loops. Language models operate by predicting the next token.  That mi...