💸Uber Burns Its Entire 2026 AI Budget by April as Agents Eat Tokens

Your AI coding bill is about to get ugly

Uber reportedly burned through its entire 2026 AI budget by April. The culprit is agentic coding. AI agents plan, generate code, run tests, and loop until they fix their own errors, and every one of those steps is billed by the token. If you're shipping with agents or paying for GitHub Copilot, this is your problem now. GitHub already moved Copilot from flat-rate pricing to metered credits. The more useful the agent gets, the more calls it makes, so your best workflows are also your most expensive ones. Cost per token tells you how much you consumed, not what you got for it. A single agentic task can use 10 to 50 times more tokens than a basic chat prompt. Gartner predicts AI coding token costs will surpass average developer salaries by 2028. Some of that spend is waste, like retries and runaway loops. Some of it is just the cost of shipping software. The fix starts with measuring cost per merged pull request and routing routine work to cheaper models you control.