Uber adds spending caps for AI coding tools after budgets could not keep up

Uber has moved from broad encouragement of artificial intelligence in software engineering to tighter cost controls, setting monthly spending caps for employees who use AI coding tools. The policy, first reported this week, is aimed at preventing another budget overrun after the company’s internal AI experiment accelerated faster than planners expected.

According to multiple reports, Uber limits each employee to $1,500 per month in token spending for each agentic coding tool. Agentic coding tools are software assistants that do more than suggest lines of code, instead taking actions like writing, modifying, and running code steps toward a task. The cap is described as applying to tools such as Anthropic’s Claude Code and Cursor, and it is tracked through an internal usage dashboard that employees can access.

The new rules come after Uber “blew through” its AI budget earlier in the year, using up the full allocation in about four months, according to reporting that cited Bloomberg and earlier disclosures. The budget depletion was reportedly not driven by reckless spending in isolation, but by adoption that exceeded expectations when developers were encouraged to use AI “as much as possible,” including through internal leader boards that ranked usage.

Uber’s leadership is also drawing attention to a second, messier issue that sits behind the budgeting problem: measurement. Even if teams can demonstrate productivity indicates at the employee level, turning those indicates into clear, organizational outcomes is difficult. Uber COO Andrew Macdonald said it is “very hard to draw a line” between reported AI-related stats and producing a higher quantity of genuinely useful consumer features.

The company’s challenge is occurring as AI coding tools become a routine part of large engineering organizations. Uber’s internal adoption is already described as widespread, with TheStreet citing figures that roughly 95% of its 5,000 engineers use AI-assisted coding tools monthly. CEO Dara Khosrowshahi has also previously pointed to AI agents contributing to a meaningful share of submitted code, reinforcing that the technology is not merely being tested in small pockets.

Uber’s decision to cap usage also fits into a broader industry shift, as enterprises look for ways to slow spending without giving up on potential gains. TechCrunch framed the move as part of a larger, unresolved question in the market: where the return on investment from AI actually lands, and whether it can be linked to outcomes that executives and investors can understand with confidence. Uber’s reports also suggest the company has moderated other plans, including hiring pace, as leaders weigh the productivity impact of AI tools.

Not everything is public yet. The reports say the caps were instituted in recent months, but Uber has not detailed how the policy affected overall AI costs, what specific engineering metrics are now being used to evaluate AI effectiveness, or how quickly the cap could change if key projects require higher usage. It also remains unclear how Uber’s cost governance will evolve if the company decides the tools are delivering more customer value than current reporting can demonstrate.