Four months. That's how long it took Uber to exhaust its entire annual AI budget in 2026. By April, the company had already hit its ceiling - and according to Andrew Macdonald, Uber's president and chief operating officer, they don't have a clean answer for what they got in return.
In an interview with Rapid Response, Macdonald said Uber isn't seeing a clear connection between rising token consumption for Claudee Code](/tools/claude-code/) (Anthropic's AI coding tool) and actual increases in developer output. More tokens burned does not equal more software shipped. That's a significant admission from a company of Uber's size, and it names a specific tool, which makes this harder to wave away as vague executive skepticism.
This isn't a fringe complaint. The ROI question around enterprise AI is becoming one of the more honest conversations in tech right now. Companies have been approving large AI budgets based on vendor promises and analyst projections rather than measured results. The problem is that measuring productivity in knowledge work is genuinely hard - you can track tokens and API calls easily, but attributing a shipped feature or a faster decision to a specific AI tool takes real instrumentation that most companies haven't built.
Uber's situation also highlights a structural problem with usage-based AI pricing. When developers have access to tools billed per token, consumption expands to fill the budget. That's not the same as value expanding with it. A developer who would have spent two hours on a task might now spend three hours prompting and reviewing AI output, burning more tokens but not necessarily delivering more.
For companies currently approving AI tool budgets for 2027, Macdonald's comments are a useful reality check. The pressure to deploy AI across engineering teams has been intense, but the accountability infrastructure - baseline measurements, productivity benchmarks, cost-per-outcome tracking - often hasn't kept pace. Uber, with more engineering resources than most, is openly saying they can't justify the spend. Smaller companies with less overhead to absorb the cost should be asking the same questions before they hit their own April wall.