In December 2025, Uber handed its engineers access to Claude Code, Anthropic's AI coding agent. By April 2026 - four months later - the company had spent its entire annual AI-tools budget. Not overshot by a little. Gone.
The story was reported by Forbes, Fortune and TechCrunch. It is the clearest public example yet of a problem quietly hitting companies of every size: AI costs that compound far faster than anyone is watching.
What actually happened
Uber rolled Claude Code out to roughly 5,000 engineers, and adoption did not creep - it jumped. According to the reporting, the share of engineers using it climbed from 32% in February to 84% by March.
- The average engineer ran $150-250 a month in tokens.
- Power users ran $500-2,000 a month.
- Multiply that across thousands of engineers over four months, and the full-year number was already spent.
Uber even ranked engineers on internal leaderboards by their Claude Code usage - which quietly turned burning tokens into a status game. Following the blowout, the company capped spending at $1,500 a month per employee, per AI coding tool.
Why AI costs compound (and headcount doesn't)
A chatbot is cheap to reason about: one question, one answer, one charge. An agent is different. It runs a loop - reads files, calls tools, makes edits, re-checks its own work - and it re-sends the entire accumulated context on every single step. By step 20 you are paying for the same instructions and history twenty times over.
That is why agentic tools consume 5-30x the tokens of a simple chat. One industry analysis put it starkly: a linear 2023-style task cost about $0.04 per run; the 2026 agentic equivalent runs closer to $1.20 - roughly 30x more for the same-looking task.
Run Uber's math on your own team
Uber's scale makes the story feel remote. The arithmetic is not. Take the per-engineer numbers from the reporting and apply them to a 20-engineer company:
- Average users: 18 engineers at $150-250 a month is $2,700-4,500.
- Power users: just two engineers at $500-2,000 adds another $1,000-4,000.
- Team total: $3,700-8,500 a month - roughly $44,000-102,000 a year, from 20 people using a tool that genuinely works.
Now add Uber's adoption curve. If you set the budget when a third of the team was using the tool and adoption jumps to nearly everyone - Uber went from 32% to 84% in a single month - the run rate roughly two-and-a-half-times itself while the budget stands still. Nobody overspends. The plan just stops describing reality. That is how a full-year budget dies in four months.
The power-user spread matters more than the average, too. In Uber's own numbers, the top of the range is about 13x the bottom: one heavy user bills like a dozen average colleagues. A team's invoice is set by its distribution, not its average - and you cannot see a distribution on a monthly invoice total.
The part everyone misses: this wasn't really about Uber
Here is the counterintuitive bit. Token prices actually fell - by around 98% over two years. And yet enterprise AI bills tripled over that same stretch, because consumption grew far faster than prices dropped. Cheaper tokens simply got used a lot more.
Gartner now projects that AI coding costs could rival what companies pay the developers themselves. Uber is just big enough to make the headline - the same curve is hitting startups and mid-size teams right now. They tend to find out on the invoice.
Four lessons that transfer to any size of company
- Budgets set during the pilot break at real adoption. Early usage is the floor, not the estimate. A budget built on month-one numbers has Uber's four-month problem baked in from day one.
- Averages hide the bill. An average of $150-250 sounds manageable right up until you meet the $2,000 engineer. Distributions predict invoices; averages predict surprises.
- Incentives move the meter. Uber ranked engineers by usage and got usage. A leadership push to "lean into AI" with no cost counterweight is a leaderboard too - just an unwritten one.
- Caps are a backstop, not a plan. The $1,500 cap is sensible, but it arrived after the money was gone. A cap without live visibility only moves the surprise to the day everyone hits it at once.
None of these lessons needs 5,000 engineers to apply. They apply the day the second person on your team opens a coding agent. And the pattern is general, not an Uber quirk: 79% of companies overshot their AI budget. Uber's version was simply big enough to report on.
What would have caught it
The failure at Uber wasn't using AI. It was flying blind. No per-engineer limits, no live attribution, no alert when the run rate went vertical. Three things turn that around:
- See it live. Spend per person, per team, per tool - updated daily, not discovered a quarter later.
- Attribute it. Know which projects and people drive the bill, so a runaway agent shows up in hours, not months.
- Cap it. Sensible per-person limits (Uber landed on $1,500) that stop one bad weekend from eating a quarter.
"We're not Uber" - true, and it won't save you
The honest objection: Uber has 5,000 engineers and dedicated platform teams, and you might have fifteen people. But scale was not the failure here - visibility was. And small companies have less of it, not more. There is no procurement review on a monthly tool habit, no analyst watching the line item, and often nobody who owns "AI spend" at all.
The mechanics - agents compounding context, adoption jumping, power users pulling the tail - work identically at 20 engineers, as the worked example above shows. The difference is what happens next. Uber could absorb the bill and turn it into policy within weeks. A smaller company just eats it. For the same failure at single-employee scale, read the Slash story: one person, one week, one very public bill.
How to check this in your own company this week
- Run the four-month test. Take last month's total AI spend and multiply by twelve. Put it next to the annual budget. If the multiplied number is bigger, you are living Uber's timeline right now - you just have not hit the wall yet.
- Dig out the budget's assumptions. What adoption rate and per-person spend was it built on? Compare with today's actuals. The gap is your blowout-in-progress.
- Chart spend per person, not just the total. Find your $150-250 middle and your $2,000 tail. Both are normal. Not knowing which people are which is the problem.
- Audit the incentives. Any mandate, metric or leaderboard that rewards AI usage with no cost line next to it deserves a second look this week, not next quarter.
- Pick the alert before you need it. Decide what number should trigger a conversation - a per-person monthly figure, a share of budget - and make sure something actually fires when it is crossed.
None of this means using AI less. Uber's engineers were getting real value from Claude Code; so are yours. The goal is simply to see the meter while it is running - which is the one thing almost nobody had in place when the bills came due.
The encouraging part: watching the meter is not a big project. Connecting usage data takes about 15 minutes, read-only, with a first report inside 48 hours and the first savings typically landing in about two weeks. That is the whole starting kit of AI cost management - in place before the invoice does the teaching, instead of after.