Walk any shop floor today, and you'll likely find a value stream map on the wall.
Walk past it on a different day, and the map still says the same thing. The shop floor doesn't.

This gap between value stream documentation and value stream reality is one of the oldest problems in Lean. It's also where most AI promises in this space fall apart. Connecting an AI agent to scattered production data does not create a working value stream view - it creates a faster way to generate incomplete answers.

What's missing is the operational context: how steps, dependencies, constraints, and losses are actually connected across the stream. Once that foundation is in place, AI agents stop guessing - and start doing useful work in continuous improvement.