Microsoft pulled the plug on its internal use of Claudee Code](/tools/claude-code/) after the AI coding assistant ran over budget, according to reports. The decision is worth paying attention to given who made it.
Microsoft sells GitHubb Copilot](/tools/github-copilot/) - its own AI coding product, built with OpenAI - to millions of developers. The fact that internal teams were running Claude Code at all suggests Copilot wasn't the universal default even inside the company that makes it. The fact that they stopped suggests the cost made the experiment hard to justify.
The Token Bill Problem
This isn't a story unique to Microsoft. Agentic coding tools like Claude Code are designed to work autonomously across multi-step tasks - reading a codebase, writing functions, running tests, fixing errors, repeating. Each step consumes tokens (think of a token as roughly three-quarters of a word). A session that a developer perceives as "one task" can silently burn through hundreds of thousands of tokens. At API pricing, that adds up quickly across a large engineering org.
Predicting those costs in advance is genuinely hard. Traditional SaaS tools have flat monthly seats. Agentic AI tools have variable consumption that spikes based on task complexity. Finance teams don't have a good framework for it yet, and the bills landing mid-quarter have caught more than one company off guard.
What This Signals
Microsoft dropping Claude Code doesn't mean Claude Code is bad - it means token economics at enterprise scale are still an unsolved problem. Cursor, which bundles model access into a flat monthly fee, sidesteps some of this pain by absorbing the cost variance itself. Tools that bill per-token pass that unpredictability directly to customers.
For Anthropic, losing Microsoft as even an informal internal user is a notable data point. Microsoft is simultaneously one of the most important enterprise AI buyers and a direct competitor through GitHub Copilot. Getting a foothold there would have been a meaningful signal. Not keeping it is the opposite.