Some developers are spending $500 a day on Claude Code. Their companies told them to.
A survey of developers at 15 companies by the Pragmatic Engineer newsletter found that spending on AI coding tools has jumped roughly 10x in the past six months. At a seed-stage startup, per-developer costs went from about $200 a month to $3,000. At a 2,000-person finance company, the initial $100-per-user monthly budget was getting blown through in three to five working days.
Many companies aren't just tolerating the spending. They're gamifying it. A senior engineering manager at a US healthcare startup said the company runs a "monthly spend leaderboard" and actively wants developers to rack up higher bills. "One of my engineers spent $1,400 on a long Claude Code session in a single day," they told the Pragmatic Engineer. "We love it." Some companies have even started rating AI usage in performance reviews, giving developers one more incentive to keep the meter running.
The math, for believers, is simple. One founder of a 700-person infrastructure company pointed out that engineers already cost $200,000 to $400,000 a year. An extra $5,000 a month on AI tools barely registers against that. The bet is that token costs will come down and productivity gains will cover the rest.
Not everyone is buying it. A director of engineering at a 10,000-person IT company offered a darker forecast: "I suspect the attitude changes once finance and other cost-conscious parts of the org realize we are spending hundreds of dollars per day, per highly-engaged developer. For now, fear of missing out keeps the taps open." At least one large gaming studio couldn't even get approval for $200 a month per developer.
GitHub just put a number on this problem. The company announced that Copilot is shifting to usage-based pricing on June 1 because the flat subscription model stopped making financial sense. GitHub's CPO Mario Rodriguez said a quick chat question and a multi-hour autonomous coding session were costing users the same amount under the old model, and GitHub had been quietly absorbing the growing gap.
To ease the transition, GitHub is offering inflated promotional credits for the first three months:
- Copilot Business subscribers paying $19 per seat will get $30 in monthly credits
- Copilot Enterprise subscribers paying $39 per seat will get $70 in monthly credits
That gap between the old price and the promotional credit tells you how much real usage had outrun the subscription.
Google is making a parallel move, rolling out new tools that give developers spending caps and usage breakdowns to help manage Gemini API budgets. You build those features when your customers keep getting blindsided by their invoices.
The frustrating part for finance teams is that per-token prices have actually dropped significantly in recent years. Total spending keeps climbing anyway because developers are using these tools far more than anyone planned for, often defaulting to the most expensive models for routine tasks that cheaper ones could handle. Deloitte recently framed AI tool costs as an entirely new budget category that most finance organizations don't have a framework for forecasting. Unlike software subscriptions, these costs scale with activity. And when you're actively encouraging everyone to use the tools more, predicting next quarter's bill is close to impossible.
GitHub's move to usage-based pricing won't be the last. Every major AI tool is heading the same direction, which means companies are about to lose the flat monthly fee that kept AI spending invisible. When every task shows up on the invoice, organizations will have to answer a question most have been dodging since they rolled out these tools. Not every $500 day of Claude Code is producing $500 of value, and the companies that can't tell the difference are going to learn that the hard way.
