Microsoft is quietly building a multi-model future. The commits prove it.

Microsoft owns 51% of OpenAI. They poured $13 billion into the partnership and restructured the deal in April to cap revenue share through 2030. And yet. Forget the press releases. Look at what Microsoft's engineering teams are actually shipping. The repos tell a different story. One where Copilot, Claude Code, and OpenAI Codex are treated as interchangeable. This isn't betrayal. It's hedging. And the evidence is public. ## The VS Code integration VS Code 1.109 didn't add Claude as some afterthought. A senior engineer on the VS Code team has been running a phased Claude agent integration, Phases 6 through 13, as primary work. Not a hackathon project. A roadmap. Microsoft added native support for CLAUDE.md files, Claude-specific slash commands (/agents, /hooks, /memory, /init), and wired in Anthropic's Agent SDK directly. Claude agent sessions now run natively inside VS Code. That's not compatibility. That's architectural commitment to a competitor. ## The plugin repos In the microsoft/ GitHub org, there's a repo where Claude Code gets first billing over Copilot in the description itself: github.com/microsoft/power-platform-skills. Then there's microsoft/win-dev-skills, built by a Principal Software Engineer at Microsoft. It ships identical plugin manifests for three platforms: Copilot, Claude Code, and OpenAI Codex. Same eight skills. Same agent. Three parallel configs. The commits tell the rest. 64 co-authored by Copilot, 4 co-authored by Claude (Sonnet 4.6 and Opus 4.7). That principal engineer personally tested the plugin with Claude Code Opus 4.7, then filed detailed issues to improve the Claude experience. The PR review skill has a cross-model verification system. Claude orchestrators pass work to GPT-5.4. GPT orchestrators pass to Claude Opus 4.7. Interchangeable. ## Why this matters right now This week, trial testimony in Musk v. Altman revealed that Microsoft internally feared "becoming too dependent on a single vendor." Even as they wrote billion-dollar checks. Executives worried about "betting the farm on a partner we don't control." The repos are the proof that those fears turned into action. Microsoft isn't abandoning OpenAI. They're building the infrastructure to not need them exclusively. Every .claude-plugin manifest, every parallel config, every cross-model verification system is a hedge. ## The pattern Microsoft has been here before. They invested in their own mobile platform while developers built for iOS. Pushed Internet Explorer while developers moved to Chrome. The difference this time: they're hedging early. Instead of fighting developer preference, they're building systems that are model-agnostic from day one. Copilot first, Claude second, Codex third. Architecturally equal. ## What to watch Whether Claude Code support lands in GitHub Copilot itself, not just VS Code. That's the line between pragmatic compatibility and strategic pivot. Whether OpenAI responds with developer tooling that competes with Claude Code directly. Codex exists but doesn't have the same gravity. Whether the jury in Musk v. Altman factors in the Microsoft dependency testimony. If they do, the "too dependent" narrative becomes a liability for OpenAI's defense.