[PLACEHOLDER — Article body goes here.]
There are two stories about AI in government. The first is the keynote story — vision decks, proof-of-concept dashboards, polished demos with synthetic data. The second is the one I actually live in: pilots that fail because the data wasn’t where it was supposed to be, procurement timelines that outlive the model architectures they were written for, and the slow realization that the hardest part was never the model.
The keynote story is not wrong. It is just not the work. The work happens in the gap between the slide and the reality, and almost nothing in the public conversation about AI in government acknowledges that gap honestly.
So this is a small attempt to close it.