AI mania is breaking corporate judgment before budgets

Simon Willison highlighted Nik Suresh’s piece about corporate AI mania, built from anonymous stories inside large customer organizations. The sharpest detail is an executive at a company with more than $2B in revenue who admitted they had never used ChatGPT or any other AI tool right after producing a technical strategy centered on AI.

Strategy is arriving before hands-on experience

Willison also quotes an engineer at a company with a token leaderboard who keeps a parallel Go repository and asks AI to rewrite it in Zig while doing other work, just to keep their job. This is not a benchmark. It is a social metric that rewards visible AI activity.

Suresh’s more useful point is vendor silence. If a customer executive talks about 100x productivity, the vendor has little incentive to correct them. Challenging the claim can look like challenging the customer’s leadership and can put an enterprise contract at risk.

Team theater is more dangerous than a weak model

The practical lesson is blunt: companies are starting to measure loyalty to AI before useful output. Token leaderboards, mandatory experiments and strategies written without personal tool experience create an environment where people learn to perform adoption instead of improving work.

For product and engineering leaders, that is the warning. AI adoption without measurements for successful tasks, review quality and time saved outside the demo quickly turns into internal PR.

Anonymous anecdotes reveal incentives, not statistics

The piece relies on anonymous examples, so it is not a market survey. Its value is in the incentive map. Anyone who has sat between a customer, an account team and an engineering team will recognize the pressure not to puncture an executive narrative.

The risk is overgeneralization. Not every corporate AI effort is theater. But when an executive sells a 100x shift without using the tool, governance has already failed before the first pilot starts.

The signal is work the team had to do anyway

The next thing to watch is less token consumption and more completed work. How many pull requests passed without extra repair? How many support tickets did AI actually close? How much manual routine disappeared outside the slide deck?

Companies that do not measure this will end up with a ritual. People will feed the dashboard, the dashboard will feed leadership and real productivity will wait in the hallway.

Lilith's verdict

An AI strategy written by people who have not used the tools is a cockpit full of passengers holding slides. The plane may still leave the runway, but nobody at the controls can admit they cannot read the instruments.