Claude Code and Codex in Copilot

[fbmcipher]

Today I found out that there's a preview feature that lets you enable and run Claude Code & OpenAI Codex within Copilot:

Somewhat confusingly, this is very different to using Claude or OpenAI models with the Copilot harness. This feature is literally running the Claude Code and Codex harnesses using Copilot's platform. So the performance and behavior of these requests should be close to, if not identical to running Claude Code or Codex on a local machine. Requests are counted and billed within normal Copilot usage limits.

As we've discussed, different harnesses can deliver very different results even when using the same model. This is all down to the deterministic behaviors of the harness, and the prompts the harness uses to guide the model's behavior.

The feature is still in preview, and there are some limitations – most notably, custom agents cannot be used when using Claude Code and Codex right now. But this could be a great way to have our cake and eat it too with the Copilot platform, especially if we ever run into intractable issues when using Copilot's agent.

Something I'll play around with and share experiences with in this thread.