Microsoft Dropped Claude Code, and the Copilot Numbers Tell the Story

The Setup

In December 2025, Microsoft quietly expanded access to Claude Code inside the company, inviting thousands of its own developers, project managers, and designers to experiment with Anthropic’s AI coding tool. The idea was straightforward: get non-engineers writing code and building things. By May 2026, those licenses were being canceled. The replacement was GitHub Copilot CLI.

The mechanics of that decision are worth thinking through carefully.

What the Adoption Numbers Actually Say

Before asking why Microsoft moved away from Claude Code, it helps to know where GitHub Copilot stands. In a recent assessment, former Microsoft executive Mat Velloso made a blunt observation: “Not even 3% of paying Copilot users use it even when it’s pre-deployed right in their faces.”

This is Copilot deployed across Microsoft’s own product stack—inside Word, Excel, Teams, Visual Studio—with zero friction to access. The usage still doesn’t materialize. That’s not a distribution problem; it’s a product-market fit problem at the behavior level.

The pattern is broader than Microsoft. Uber reportedly burned through its entire 2026 AI tooling budget in four months. Enterprises everywhere are experiencing the same gap: procurement of AI tools is fast, genuine workflow integration is slow, and the two numbers look nothing alike on a quarterly review.

Why Microsoft Still Dropped Claude Code

The cancellation is almost certainly not a judgment on Claude Code’s capability. The tool works. Microsoft employees who were using it presumably adopted it because it was useful, not because it was mandated. Pulling a tool developers like, right when you’re trying to convince people that AI is transforming how work gets done, is a strange move on the surface.

The more credible explanation is structural. Microsoft owns GitHub. GitHub owns Copilot. Claude Code is an Anthropic product. Using a competitor’s coding tool internally—at scale, with thousands of licenses—is awkward market messaging for a company betting heavily that Copilot is the AI development platform.

There’s also the financial angle. Microsoft restructured its OpenAI partnership earlier this year, with the cloud exclusivity arrangement ending in April. The company is recalibrating where AI spend goes and who captures the value from it. Paying Anthropic for internal licenses while promoting GitHub Copilot externally is exactly the kind of line item that gets flagged in a cost review.

GitHub Is Carrying a Heavy Load

Tom Warren’s investigation at The Verge described GitHub as “fighting for its survival” within Microsoft. That framing is pointed—GitHub’s platform business, measured on its own terms, is healthy. The question is what metric GitHub is now judged by.

When you acquire a developer platform for $7.5 billion and then build an AI coding product on top of it, the original product’s success becomes inseparable from the new product’s adoption. GitHub’s standing inside Microsoft is now tied to Copilot’s adoption curve—and those numbers are not where they need to be.

The tension is visible in how Copilot has evolved: deployed aggressively, on every Microsoft surface, and still underused. Meanwhile, tools with more focused surface areas—Claude Code, Cursor, Windsurf—are capturing developer attention precisely because they’re not trying to be everywhere at once.

What This Means for Teams Choosing Tooling

For teams making real tooling decisions right now, the Microsoft story carries a few useful signals.

Enterprise AI tool decisions are increasingly political. When a vendor cancels a tool their own developers prefer, they are optimizing for something other than developer productivity. Recognizing that dynamic helps you evaluate vendor choices with clearer eyes.

The tools developers adopt out of genuine preference and the tools enterprises procure to satisfy an AI strategy are increasingly diverging. That gap exists at Microsoft, and it probably exists in some form at any large organization deploying AI tools at scale.

For smaller teams—startups and scale-ups working at the pace we do at Dracode—the absence of internal politics is a genuine advantage. You can simply pick the tool that makes shipping mobile products faster and move on. When a better tool arrives, you switch.

The 3% Copilot engagement number is striking, but it points to something most engineering leaders already sense: deploying a tool and having people use it are solved by entirely different things. Deployment is an IT problem. Adoption is a workflow design problem. Most enterprise AI rollouts spend heavily on the first and almost nothing on the second.

What We’re Watching

GitHub Copilot’s next few months will be clarifying. Either Microsoft finds a way to close the engagement gap—through better product design, tighter workflow integration, or simply more time—or the tools developers actually reach for start receiving more institutional support rather than less.

The Microsoft-Anthropic dynamic is worth watching too. Even as Microsoft cancels internal Claude Code licenses, Anthropic just signed a compute deal with SpaceX and is scaling capacity aggressively. If Claude Code keeps widening its capability lead on long-horizon agentic tasks, the tension between what’s strategically mandated and what developers actually want will only get harder to paper over.

Sources

1. Microsoft starts canceling Claude Code licenses — The Verge, May 14, 2026
2. Microsoft cancels Claude Code licenses, shifting developers to GitHub Copilot CLI — Windows Central, May 15, 2026
3. GitHub faces a fight for its survival at Microsoft — The Verge, May 21, 2026
4. Ex-Microsoft exec says the company blew it with AI — Windows Central, May 19, 2026
5. OpenAI and Microsoft’s alliance fractures as cloud exclusivity deal ends — Tom’s Hardware, April 28, 2026