Meta and Microsoft are pulling staff off Claude, and engineers lose the choice of tool

Rswebsols.com reported this week, citing a 5 October report by The Information, that Meta and Microsoft have begun cutting employee use of Anthropic's Claude in favour of their own coding tools. Microsoft had expected internal spending on Anthropic technology to pass $1 billion a year. That estimate has fallen by more than a third since management told staff to use Claude less and use GitHub Copilot and OpenAI models more. At Meta, Claude Code users fell from around 60,000 earlier this year to about 30,000.

The employer chooses the tool

Engineers at both companies built daily habits around one assistant. Then management changed the assistant. In Microsoft's cloud and AI division, monthly AI spending limits fell from $100,000 per employee to about $10,000 in most cases, according to the report. The report describes those figures as spending ceilings. Even so, the report says the cut disheartened engineers who had been free to try different models.

The reason is commercial. Both companies are major Anthropic customers, and both compete with it. Meta's replacements, MetaCode and Muse Code, run on Meta's own models. MetaCode has more than 30,000 internal users and Muse Code more than 6,000. Meta began testing Muse Code with external clients in August. The report says layoffs played a part in the fall in Claude Code users, but the main cause was Meta's move to its own tools.

Microsoft, meanwhile, keeps selling Claude. Customer spending on Anthropic models through its enterprise platforms is still rising, the report says. So the company sells this model to customers while steering its own staff away from it.

This was a procurement decision.

Fewer users, a large bill

The number of users fell while the bill stayed large. During the transition, Meta reportedly spent more than $105 million on Claude Code over a 28-day period. Anthropic's annualised revenue is reported to be running at $65 billion, so overall demand is holding up.

Taken with Microsoft's budget caps, the signal is plain. At two substantial Anthropic clients, AI spend per employee is now a line item someone manages. Open-ended experimentation is giving way to spending tied to tools the company owns.

Changing assistant does not remove the risk. Cybersecurity News previously reported flaws in Claude Code that could have allowed unauthorised execution and API key theft, which Anthropic fixed before they were made public. Microsoft has had similar problems. The RoguePilot vulnerability, now fixed, showed how malicious commands hidden in a GitHub Issue could be used to take over a repository. Whichever tool the company picks, a person still has to check what it does.

AI does the routine work. You do the thinking. The engineers who come through this well are the ones whose value sits in the thinking, so it moves with them from one tool to the next.

Your Next Move

1. Make your method portable. Write down how you work with an assistant: how you break down tasks, the prompts you rely on, and your review checklist. Keep it separate from any one product. This month, run the same task through a second tool and note what changes. If your employer swaps the tool, the adjustment should take days, and your edge should come with you.

2. Put a number on what your AI spend produces. If a ceiling can drop from $100,000 to $10,000, the people who keep their freedom are the ones who can show what a month of usage delivered: features shipped, defects caught, hours returned to the team. Our guide on how to show impact when your employer grades your AI use sets out how to record it before your next review.

3. Own the review step. Set tight permissions, use trusted project settings and read AI-generated actions before they run. Whatever the procurement team chooses next quarter, that judgement stays with you.