GitHub Copilot

GitHub Copilot isn't just an autocomplete box anymore. It's closer to a dispatch layer that sits on top of your repositories. You pick from what GitHub calls "leading LLMs optimized for speed, accuracy, or cost," then hand the work to an agent: Copilot itself, your own custom agents, or third-party ones like Claude by Anthropic and OpenAI Codex. That work can start almost anywhere. GitHub lists the IDE, the CLI, GitHub itself, project tools, chat apps and custom MCP servers as places Copilot runs, and there's a separate MCP Registry for wiring in external tools.

The core loop is issue to merge. You assign a task, the cloud agent will "plan, explore, and execute work autonomously in the background," and you review the resulting changes before merging. The new GitHub Copilot desktop app is built to manage that from one place: launch work from GitHub, track progress across multiple agents, review diffs and merge. Inside your editor, Copilot explains concepts, completes code, proposes edits and validates files through agent mode. To keep agents from guessing, Copilot Spaces lets you build a shared source of truth from your docs and repos, so the model works from your project's context and not just the open file.

The pitch only fully lands if your code already lives on github.com. The tight coupling to Issues, Code Review and the repo itself is the whole advantage, and it's what a standalone AI editor can't easily copy. The Free plan is a real trial, not a teaser: 2,000 completions a month, access to Haiku 4.5 and GPT-5 mini, Copilot CLI, and no credit card required. Paid plans start at $10/mo. For teams, the governance story is unusually concrete: audit logs, a single control plane for agents, and allow lists that control which MCP servers developers can reach from their IDEs. Students who verify get a dedicated Copilot Student plan.