Seven companies that have pulled back on AI recently:
@Uber → burned its entire 2026 AI coding budget in four months
@Microsoft → canceled most of its internal Claude Code licenses when the token bill blew up
@github → moving Copilot off flat-rate to per-token billing
@cursor_ai → scrapped its "unlimited" plan after usage bills spiked
@Klarna → rehiring humans after its AI support quality dropped
@CommonwealthBank → rehired 45 staff, called the AI cuts an "error"
@duolingo → pulled AI back out of employee performance reviews
Every one launched chasing the same thing: PRODUCTIVITY. But productivity is subjective and hard to pin down, so companies measured what was easy to see instead. USAGE.
Microsoft, Meta and Shopify started scoring people on AI use in reviews. Amazon ran an internal token leaderboard. Nvidia's CEO said he'd be "deeply alarmed" if a $500k engineer wasn't burning $250k in tokens a year.
So people gamed it, running AI on everything to pump their numbers. Tokenmaxxing.
Except usage was never output. Jellyfish data has the cost per merged pull request rising from $0.28 with light AI use to $89.32 with heavy use. More tokens, not more shipped. Just a bigger bill.
And the bill was the one number that always came in clean. On time, to the cent, every month.
That's the streetlight effect. We count what's lit and treat it as the thing we actually care about.
But people don't keep spending $2,000 a month on something that isn't working. The value was real. It just doesn't land where the cost does. You feel it while you're using the tool. The bill lands on the company's books. So the company makes the call.
Uber’s COO has said that it’s getting “harder to justify” its AI costs because there was no way to show a link between AI spend and any meaningful increase in useful features. This is the first time I’ve seen a company say this directly.
https://t.co/xUhZvtpwah