Microsoft is canceling Claude Code for its engineers. They have until June 30 to switch to Copilot CLI.

Microsoft began revoking internal Claude Code licenses on May 14, with all engineers in its Experiences + Devices group expected off the tool by June 30, The Verge’s Tom Warren reported. The cutoff lines up with the last day of Microsoft’s fiscal year.

The group that’s losing access isn’t a small team. Experiences + Devices is the division that builds Windows, Microsoft 365, Teams, Outlook, and Surface, and it represents thousands of engineers who picked up Anthropic’s CLI when Microsoft opened access in December. Six months in, the tool got popular enough internally that it eclipsed GitHub Copilot CLI, the product Microsoft has been quietly trying to make engineers use. The official framing is “toolchain unification.” The timing says cost.

What changed and what didn’t

Microsoft EVP Rajesh Jha defended the move in a statement picked up by TechRadar: “Claude Code was an important part of that learning. At the same time, Copilot CLI has given us something especially important: a product we can help shape directly with GitHub for Microsoft’s repos, workflows, security expectations, and engineering needs.”

A few things sit underneath that.

• The Anthropic partnership isn’t ending. Claude Opus 4.6 and 4.7 stay accessible through Microsoft Foundry, and Anthropic models stay routable inside Microsoft 365 Copilot. What’s being cut is Anthropic’s tool, not its model. Copilot CLI itself still calls Claude alongside OpenAI and Gemini.
• The fiscal-year timing is the giveaway. Windows Central reported the move is “likely driven by financial motives,” and that’s consistent with the timing. Agentic coding tools burn tokens. The Verge cites engineers who say full annual AI budgets got consumed in months under token-based billing, the same pricing structure that pushed GitHub Copilot to a 27x multiplier on Claude Opus for retail customers in May.
• Engineers don’t get a vote. Tom Warren’s reporting describes engineers who picked Claude Code over Copilot CLI and are being told to switch anyway. The product Microsoft can govern beat the product Microsoft sells, and Microsoft isn’t going to let that ride into a second budget cycle.

What this looks like from the Microsoft side

The cleanest read is that Microsoft has two AI dev tools on its books and only one of them generates margin. Claude Code is a paid subscription to a competitor’s product. Copilot CLI is Microsoft’s own surface, built on Microsoft’s own infrastructure, with Microsoft’s own pricing on top. Even if Anthropic’s models stay in the loop, every token a Microsoft engineer spends through Claude Code is a direct cash transfer to Anthropic. Every token they spend through Copilot CLI is a captive workflow Microsoft can monetize and audit.

The internal-popularity problem makes that worse. If engineers prefer Claude Code’s UX and Microsoft’s own tool is the one that needs the user research, you have a vendor-lock-in problem inside the building that builds the vendor. Killing the popular alternative is the cleanest fix, and it sends the signal externally: Microsoft is committed to Copilot, even when its own people aren’t.

It’s also worth reading this against VS Code’s brief experiment of stamping every commit with Co-Authored-by: Copilot by default. The same posture shows up: Microsoft wants Copilot to be the default surface, even when the developer experience pushes back.

What this means for you

If you’re a Microsoft engineer in Experiences + Devices, you have five weeks to port your workflow. Copilot CLI now ships Claude as one of three model backends, so the model swap isn’t the hard part. The interface, the agent loop, and the shell integration are where the muscle memory breaks. Treat June as a forced migration window.

If you’re not at Microsoft, the lesson is sharper. The largest enterprise customer Anthropic’s tool had just got pulled, not because the tool stopped working but because the vendor it competes with happens to write the checks. Plan AI-tool adoption with the assumption that one product owner can unwind years of buy-in in a single fiscal-year decision. Run a model layer your team controls, and treat the agentic shell on top as swappable. The tool that wins on the engineer’s laptop doesn’t always win on the EVP’s spreadsheet.