Uber Caps Usage of AI Tools Like Claude Code to Manage Costs

Simon Willison, relaying a Bloomberg report, notes that in order to manage AI tool costs, Uber has in recent months set spending caps on employees' use of agentic AI coding software. According to an Uber spokesperson, every employee's monthly token spending cap on each individual AI coding tool is $1,500; in other words, usage of one tool does not affect the quota of another. This limit applies only to agentic coding software like Cursor and Anthropic's Claude Code, not to all AI tools. Willison interprets this within a larger context: he had previously mentioned that Uber used up its AI budget in the first four months of 2026, and he considers this unsurprising, because the relevant budget was likely set in 2025, when enterprises could hardly predict that "token-burning" coding agents would become so popular. To him, a cap of $1,500 per month per tool is more reasonable than using tokenmaxxing leaderboards to encourage employees to competitively increase AI usage, and it is a pragmatic response to overspending. The article also uses this figure to estimate the value enterprises place on AI coding tools: if an engineer aggressively uses two tools, the combined monthly cap is $3,000, making the annual cap $36,000. Compared with the median total annual compensation package of around $330,000 for Uber software engineers in the U.S. on Levels.fyi, the annual AI tool spending cap accounts for about 11%. Willison also adds his own usage situation: he currently consumes roughly $1,000 worth of tokens per month each on Anthropic and OpenAI, but because his personal subscription plans are subsidized, he actually pays about $100 to each; such discount plans do not apply to large enterprises like Uber. If he worked at Uber under his usage pattern, he would still have about $500 of monthly quota available per tool.