Uber slammed the brakes on its AI spending after blowing through its 2026 budget in just four months. The rideshare giant now caps employee monthly spending on AI coding tools at $1,500 per platform. This limit applies to agentic AI systems like Claude Code and Cursor, which write and modify software code with minimal human input.

The company’s AI enthusiasm ran ahead of its wallet. Internal leaderboards once ranked employees by their AI usage, encouraging heavy consumption. The result: Uber exhausted its entire yearly AI budget by April. The new policy splits spending limits by tool, so using multiple platforms multiplies an engineer’s allowance. Employees track usage on internal dashboards and can request approvals to exceed limits.

This move reflects a broader industry challenge: AI can boost productivity but carries unpredictable, variable costs. Agentic AI tools charge per token or inference, meaning costs balloon as engineers run complex, multi-step workflows. Uber’s cap translates to roughly $36,000 per engineer per year on AI—a sizable slice of their median $330,000 compensation.

Uber’s leadership openly acknowledges the tension between AI’s promise and its price. CEO Dara Khosrowshahi revealed that about 10% of the company’s code submissions now come from AI agents. Yet, COO Andrew Macdonald admitted on a podcast that linking AI usage directly to new consumer features remains difficult. Productivity gains are visible, but the long-term business value is still murky.

The restrictions come amid a growing corporate trend. Other tech giants, including Microsoft, have imposed similar controls after facing soaring AI expenses. Microsoft is shifting some teams away from agents like Claude Code toward alternatives like GitHub Copilot CLI to manage costs. These spending controls—rate limits, dashboards, and approval workflows—are becoming standard governance tools for enterprise AI.

Uber’s experience underscores a harsh reality: AI adoption at scale demands discipline. Experimentation drives innovation but unchecked token consumption will wreck budgets. Companies must balance enthusiasm with cost controls to avoid bleeding money on AI without clear returns.

As AI tools embed deeper into workflows beyond engineering—spreading to legal, marketing, and other functions—containing expenses grows more urgent. Uber’s spending cap is a blunt but necessary instrument. It curtails runaway costs while preserving room for meaningful AI experimentation. Until better ROI metrics emerge, firms will rely on such guardrails to keep AI spending from spiraling out of control.