Uber’s AI Spending Story Is Really About Measurement

Uber rolled out Claude Code in December 2025, and use of agentic coding tools among its engineers rose from 32% in February to 84% in March. [1] During that period, the company exhausted its entire 2026 AI budget in four months, a figure CTO Praveen Neppalli Naga confirmed to The Information in April. [1] In June, Bloomberg reported that Uber responded with a hard cap of $1,500 per employee, per month, for each tool. [1]

COO Andrew Macdonald said it was “very hard” to connect those usage statistics to producing 25% more useful consumer features. [1] The CIO.com column argues that Uber’s central problem was measurement rather than token prices because the company could not connect spending to what the tools delivered. [1] The author says the AI-cost denominator is empty at most companies, making proven KPI improvements insufficient when finance cannot establish that returns exceed costs. [1] The proposed ROI exchange rate assigns dollar values to KPI movement before deployment, such as one point of first-call resolution or one recovered engineering hour. [1]

Uber’s experience separates spending controls from value measurement: a $1,500 monthly tool cap can contain costs, but it does not show whether Claude Code produces enough valuable output to justify them. The operational requirement is to define how improvements such as first-call resolution or recovered engineering time convert into dollars, then record tool costs against that value. Without both sides, even adoption rising from 32% to 84% cannot settle the ROI question. What the sources don't address: whether Uber subsequently established a measurable link between agentic coding use and useful consumer features.