Uber burned through its AI budget months into 2026. That's $3.4 billion gone, and CTO Praveen Neppalli Naga says the company is "back to the drawing board." The culprit? Usage of AI coding tools, particularly Claude Code, that "far exceeded internal expectations."
Uber created internal leaderboards to track how much engineers used tools like Claude Code and Cursor. Engineers responded with what Hacker News commenters dubbed "token maxxing," inflating token consumption to climb the rankings. Now 11% of Uber's live backend code updates are AI-generated, handling ride-matching and pricing.
Nobody's shown the math on whether this saves money.
More AI-written code isn't automatically better code. Look at Uber Eats menu descriptions, which appear AI-generated with generic, repetitive text that probably doesn't drive orders. If that's the visible output, what does the backend code look like six months from now?
The real failure is leadership's obsession with usage metrics over outcomes. Pushing engineers to burn tokens for leaderboard standing isn't a strategy. It's a budget hole dressed up as innovation. This highlights the challenge many companies face, a situation driving a market shift where investors are ghosting OpenAI in favor of Anthropic.