GitHub

June 3, 2026 - Link Blog

Uber limits the use of AI tools such as Claude code to manage costs. I wrote that Uber blew its 2026 AI budget in four months, which wasn't all that surprising considering it budgeted for 2025 before anyone could predict how popular token-burning coding agents would become.

Bloomberg's Natalie Lung:

An Uber spokesperson said in response to a Bloomberg News investigation that the ride-hailing giant is capping monthly token spending per AI coding tool for every employee at $1,500. This means that spending on one tool will not affect your budget for another tool. The restrictions enacted in recent months only apply to agent coding software such as Cursor or Anthropic PBC's Claude Code.

A cap of $1,500 per month per tool strikes me as a reasonable policy response to overspending, and much more reasonable than a token-maximizing leaderboard that encourages employees to compete to use AI as much as possible.

It's also interesting that it hints at a real dollar value for the value Uber gets from these tools. Assuming two actively used tools per engineer, this results in a limit of $3,000 * 12 = $36,000 per engineer per year. Levels.fyi lists the average annual compensation package for Uber software engineers in the United States as $330,000.

This means that each employee's AI spending limit is ~11% of their median compensation package.

I noted that my token usage for Anthropic and OpenAI each amounts to about $1,000 per month. Currently, the cost per provider is just $100 thanks to generous subsidy plans for individual subscribers. These plans are no longer available from larger companies like Uber.

Their new policy means that if I were working at Uber, I could set aside ~$500 in tokens per month for each tool, given my current usage patterns.