Uber Burned Its Entire 2026 AI Budget on Claude Code by April

Uber ran out of money. Not the company. The AI tooling budget.

Briefs reports that Uber exhausted its annual AI tooling budget by April 2026, four months into the year, after rolling out Claude Code to its engineering org in December 2025. Usage doubled by February. Individual engineers racked up $500 to $2,000 per month in API costs. Against a $3.4B total R&D budget, the CTO said they're "back to the drawing board" on AI budgeting.

The story hit 397 points with 469 comments on Hacker News. That comment count tells you this isn't just an Uber problem. It's every engineering leader's problem.

Here's what happened: Uber gave engineers access to the best coding tool available, those engineers used it because it made them faster, and the aggregate spend blew past every forecast model the finance team had. Nobody planned for 95% adoption. Nobody modeled what happens when a tool is so useful that developers voluntarily use it for everything. The traditional enterprise software playbook assumes friction limits adoption. AI coding tools have negative friction. They're addictive.

I've been tracking this pattern since the $6,000 /loop incident last week. But that was one developer making a mistake. Uber is an entire engineering organization making rational individual decisions that collectively overwhelm the budget. Each engineer spending $1,000/month is getting enormous value. The problem is 5,000 engineers doing it simultaneously.

The 70% AI-generated committed code stat is the one that should scare CFOs. If seven out of ten lines touching production came through Claude Code, you can't just turn it off. You've built a dependency. Cutting the tool means cutting productivity. Keeping it means finding budget that doesn't exist.

This is the enterprise AI cost story of 2026. Not "AI is too expensive to use" but "AI is too valuable to limit." Those are very different problems, and the second one is harder to solve.

What to do about it: If you're managing an engineering budget, build your AI tooling forecast on 80%+ adoption, not the 30-40% pilots suggested. Use the proxy coworker pattern (story #2 below) to route cheap tasks to cheap models. And talk to your CFO now, not after you've burned through Q3's allocation.