The story has been making the rounds, and it is a good one. Uber rolled out Claude Code to about five thousand engineers in December 2025. By April, four months later, it had spent its entire AI budget for the year. Not overspent by a little either — the entire allocation was simply gone.
It is a remarkably clean illustration of a shift that a lot of technology teams are feeling right now.
What actually happened
Here is what actually happened, because the mechanics matter. Engineers loved the tool almost immediately. Adoption jumped from 32% to 84% in a matter of weeks. Heavy users ran up 500 to 2,000 dollars a month, and the CTO himself spent 1,200 dollars in a single two-hour session. Uber’s overall R&D spending was not really the problem here; the underlying pricing model was.
Flat fees are giving way to consumption pricing
For years, software was priced by the seat. One engineer, one license, one predictable line item you could multiply by headcount. Consumption pricing does not work that way. The same engineer, on the same day, can produce wildly different bills depending on what they ask the tool to do. Microsoft caps its Copilot at a flat 30 dollars per user. A token-based tool gives finance almost no forward visibility at all.
The answer is not to pull back
So the natural instinct is to pull back, and that is almost always the wrong move. Remember why the budget blew up: the tool was too useful to put down. At Uber, roughly 70% of committed code came from AI. Pulling back would forfeit real gains to avoid a forecasting problem. The answer is not less AI, but scaling AI more deliberately.
Start small — with a path to production
And deliberate really is the operative word here. A slick demo can feel like a win, but real return shows up at scale, embedded in your daily workflows, not in the proof-of-concept. Start small only works when it is paired with two things: a clear path to production, and cost guardrails built in from day one. Bolted on after a budget blowout has already happened, guardrails are really just expensive cleanup.
Which use cases matter most?
That is why the first question matters more than the tool. Which use cases are actually critical to your business? Begin with the high-value, low-risk ones that carry the least downside. Prove them out properly, then expand into the processes where AI clearly pays off, with the guardrails already in place.
A full-year budget, in four months
We’re solutions people
This is the work we focus on. We are solutions people, so every engagement is built for how your business actually operates, never a template forced onto you. We help you map the use cases that matter, put the guardrails in early, and scale where the return is real, across Salesforce, ServiceNow, and the custom workflows in between. As your partner, we own the outcome with you.
Your AI bill should never outrun the return it creates, and with a clear plan in place, it does not have to. If you want the bigger picture on why depth of use, not model choice, is what pulls firms ahead, we covered it in the AI frontier gap.