The best tool you're not allowed to keep

Microsoft is cancelling its Claude Code licences and shifting developers to GitHub Copilot CLI. Windows Central's framing points to financial motives rather than developer satisfaction. But the interesting reason is what it implies: Microsoft would rather ship a tool it owns than subsidise one it doesn't, regardless of what engineers actually prefer.

The vertical integration play

This is a pattern economists would recognise immediately. When a capability moves from nice-to-have to core workflow, the platform owner pulls it inside the wall. It happened with browsers, app stores, and cloud databases. Now it's happening with coding agents. The moment your AI assistant touches every file in your repo, runs arbitrary shell commands, and shapes architectural decisions, "which vendor" becomes a question about who holds the keys to your engineering velocity.

The same day Microsoft locked the door, xAI kicked one open. Grok Build enters the terminal agent race as a new competitor. OpenAI went the other direction, making Codex available from anywhere so developers can interact with their codebases remotely. Two different bets on the same insight: the coding agent is becoming the interface through which engineers interact with their entire codebase, and whoever owns that interface owns an extraordinary amount of leverage.

Meanwhile, PwC and Anthropic are expanding their alliance for enterprise agentic AI. When a Big Four firm goes deep on a single AI provider, it's a signal: the coding agent is no longer just a developer productivity tool. It's becoming institutional infrastructure.

What the preference gap reveals

The way I see it, the real story is that developer preference and developer tooling have been decoupled. For a brief window, individual engineers could pick their own agent the way they once picked their text editor. That window is closing. The choice is moving up the org chart, from the person writing code to the person signing contracts.

This creates a genuinely new tension. Coding agents have different models, different context handling, different failure modes. A team that's built muscle memory around one agent's idioms doesn't swap to another without friction. But the platform owner has reasons beyond quality to pull the tool inside the wall.

For builders and engineering leaders, the practical question is sharpening: do you optimise for the best agent today, or for the platform that will still let you use it in eighteen months? PwC's answer is to go deep with one provider. Microsoft's answer is to own the stack regardless of current quality. Both are rational. They just optimise for different things.

The coding agent started as a productivity hack. It's now a platform control decision dressed in a developer tool's clothing. Pick accordingly.

---