The AI Buffet Now Comes With a Bill

Imagine your manager stops by your desk and says: coffee is still free, but only the first two cups. After that, please put it on a project code.

That is roughly what happened at Accenture this month. Except the coffee is Claude.

Twenty-nine dollars a month

According to an internal memo first reported by 404 Media, Accenture now centrally funds $29 per employee per month for Claude, starting October 1. Anything above that has to be charged to a project or team budget through a WBS, the company's internal billing code. The memo also says that getting a license and a budget code "does not guarantee indefinite or ongoing access to Claude." At least one employee reportedly lost access completely.

The reason is not hard to guess. Employees quoted in the reports said their usage ran between $500 and $1,000 a month, mostly on coding work. Accenture has around 814,000 people. You can do that math yourself.

The official answer from a spokesperson was short: "Our approach today is to provide the right tool for the right work."

A small, uncomfortable question

What did your AI usage cost last week? Not the subscription. The actual usage.

I'd bet most engineers can't answer that. And for a long time that was fine. Companies wanted adoption, and you don't get adoption by showing people an invoice on day one.

But a hidden price changes behavior. When something feels free, we use it for everything. We let an agent "look around the codebase" for twenty minutes to rename a field. We rerun the same failing prompt five times instead of reading the stack trace once.

We have seen this movie before. It was called the cloud. First everyone spun up whatever they needed, then the bill arrived, and suddenly there were tags and team budgets. AI is walking the same road, just faster.

Free was never a price. It only meant someone else was reading the bill.

Where this is heading

I don't think Accenture will be the last. Maybe others won't pick $29, and maybe they won't be this blunt, but per-team AI budgets and the question "which project pays for this?" are coming to a lot of companies.

There is a business detail in the coverage that explains a lot. According to 24/7 Wall St., more than 65% of Accenture's bookings are fixed-price work. On those contracts, every extra token comes out of Accenture's own margin unless it was priced in. If you build software on fixed-price projects, this is your problem too, whether anyone has said it out loud yet or not.

And there is a trap here for managers. A flat cap set at the level of a consumer subscription hits hardest the people who use AI most seriously. Your heaviest users are not always your most wasteful ones. Sometimes they are your best ones.

Cost awareness is a senior skill

A good senior engineer already knows which query is expensive and why a Friday deploy is risky. Tokens now belong on that list. Knowing what your work costs is not about being cheap. It's about being trusted with more.

What we can do now

If you write code:

1. Look at your usage once a week. Most tools show it somewhere. Just knowing the number changes how you prompt.
2. Match the tool to the job. Not every question needs the biggest model or an agent running for twenty minutes.
3. When a session starts going in circles, stop. Read the error yourself, then start again with a clearer prompt. It's usually cheaper than the tenth retry.

If you lead a team:

1. Make costs visible before you cap them. People improve what they can see. Surprise limits mostly create frustration and shadow tools.
2. Tie spending to outcomes, not to headcount. A flat number per person treats your power users exactly like the people who never open the tool.
3. Decide who owns the AI budget now, before the first big invoice decides for you.

The buffet isn't closing. It's just getting a menu with prices on it. I think that's healthy. The engineers who learn to read that menu early will be the ones handed a bigger one.

Sources: 24/7 Wall St. via Yahoo Finance: Accenture just capped employees' Claude spending at $29 a month, HR Katha: Accenture caps central funding for Claude AI at $29 per employee per month