Measuring real AI adoption needs stable baselines

Tracking enterprise AI adoption over time requires consistent questions asked to the same demographic cohorts. When Simon Willison revisited the long-running U.S. Census Bureau survey on company AI use, an executive order centered on super intelligence had complicated the government's tracking. The Census data had been one of the few neutral datasets available, used notably in an Apollo Global Management analysis that mapped an adoption slowdown among large firms in late 2025.

The trouble with importing speculative terminology into federal statistical collection is that it obscures practical enterprise reality. Most corporate deployments are not pursuing speculative intelligence thresholds; they are wrestling with document retrieval, data governance, and integrating models into relational databases. Framing official inquiries around extreme capabilities makes it harder to measure where software teams are making genuine progress and where adoption has stalled on integration friction.

For teams budgeting multi-year technical roadmaps, reliable public baselines matter because they separate durable adoption from vendor marketing. When national statistical agencies bend operational tracking toward policy slogans, engineering leaders lose the clearest independent signal they had on whether enterprise usage is expanding or hitting a ceiling.