Uber exhausted its entire 2026 AI budget by April after rolling out Claude Code and Cursor in December 2025, with per-engineer API costs hitting $500-$2,000/month.
Key Takeaways
Claude Code launched internally in December 2025; usage doubled by February, with 95% of engineers using AI tools monthly and 70% of committed code AI-originated.
Cursor plateaued while Claude Code dominated; CTO says the company is “back to the drawing board” on AI budgeting.
Per-engineer monthly API spend of $500-$2,000 across ~5,500 engineers implies roughly $2.75M-$12M for four months – a small slice of Uber’s $3.4B R&D budget.
Leadership framed limiting access as counterproductive, but has not published productivity ROI data to justify continued spend at scale.
Hacker News Comment Review
The core skepticism: commenters widely rejected the article’s assumption that high spend equals high value, noting that burning a budget proves engineers like the tool, not that it ships better products or grows revenue.
Several engineers questioned how $1,000+/month in tokens is even achievable responsibly; one reply attributed it mainly to beginner habits like reusing long-lived conversations without summarization checkpoints, inflating context costs.
A commenter ran the math and found annualized AI tool spend is roughly 1% of Uber’s R&D budget, undercutting the “meaningful chunk” framing and suggesting the real failure was forecast modeling, not the tools themselves.
Notable Comments
@ninjagoo: Back-of-envelope: 5,500 engineers at $1,250 midpoint = ~$6.8M over four months, or ~1% of R&D annually – not the significant budget share the article implies.
@MichaelNolan: “70% of committed code originating from AI” is expected when AI tool usage feeds into performance evaluations.
@tills13: AI is a force multiplier regardless of direction – engineers with poor fundamentals can use it to create messes faster.
