If you’re using AI to write code, I’d stop letting the same agent be the only judge of its own work. GitHub just made Copilot code review callable through an API. That makes it much easier to build a workflow like this: Builder → Reviewer → Fix → Reviewer → Merge 1 - Your coding agent builds the feature. 2 - A separate AI reviewer inspects the pull request. 3 - It reviews the code for issues and checks it against your repository’s rules and context. 4 - The findings get sent back into the coding workflow to be fixed. 5 - The reviewer checks the new changes again. 6 - Only then does the code move toward merge. You can even give the reviewer its own skills and MCP context to improve how it reviews your code. The important part isn’t Copilot. It’s separating creation from verification. As coding agents get faster, generating more code probably won’t be the bottleneck. Trusting the code will be. So instead of only building increasingly powerful coding agents, I’d start building independent verification around them. One agent builds. Another tries to prove it wrong.