Uber's AI agent use is up 9.4x while spending flatlines
Two enterprise signals landed on the same day: Uber flattened its AI bill even as agent use exploded, and Workday showed AI agents now pulling customers into core platform deals. Both point the same direction — enterprise AI is moving from a cost headache to something a CFO can actually budget.
Uber says weekly AI agent requests have grown 9.4x since February, while its total AI spending has stayed flat since April. The company burned through its 2026 AI budget in the first quarter, so it built cost controls instead of cutting access: a routing layer sends simple tasks to cheaper models and reserves frontier models for hard jobs, sessions are capped at 400,000 tokens even when the underlying model supports a million, engineers see the live cost of a session in their terminals, and the prompt cache was extended from five minutes to an hour. Employee adoption of AI tools rose more than fourfold over the same stretch, Uber told Axios. Read as a playbook, this is the problem every enterprise is about to hit — agent usage scaling while the finance team insists the bill not follow it.
Workday beat expectations and said AI is now doing the selling: more than half of its net new customers signed up for at least one AI solution last quarter. Revenue rose nearly 13% to $2.65 billion, subscription revenue climbed 13.9% to $2.47 billion, and AI annual recurring revenue reportedly jumped 200% to nearly $600 million. Analysts had spent a year worrying that generative AI would undercut legacy enterprise software; instead Workday's commercial chief says the AI pitch is strong enough to pull customers into full platform switches — what one analyst called the "opposite of the SaaSpocalypse." The shares whipsawed, falling as much as 5% before recovering in extended trading as investors weighed an in-line forward forecast.
What to watch: whether Workday's AI ARR keeps compounding and whether Uber's cost controls hold as agent volume keeps climbing.
Is your company's AI bill scaling faster than its AI usage? Tell us in the comments.
