- GitHub’s latest Copilot agent is embedded straight into the platform
- It’ll boot a secure dev environment and clone your repo before cracking on
- If you need to make further changes, just leave a comment in the thread
Announced at Microsoft‘s annual developer conference, Build 2025, GitHub launched a new and updated version of its Copilot AI assistant designed to streamline the integration of computer-aided coding even further.
“GitHub Copilot now includes an asynchronous coding agent, embedded directly in GitHub and accessible from VS Code,” the company wrote.
GitHub CEO Thomas Dohmke explained how the agent gets to work in the background when you assign a GitHub issue to Copilot or prompt it in VS Code, adding that it enhanced productivity without putting organizations’ security at risk.
GitHub’s Copilot agent sits quietly in the background, ready to spring into action
“Having Copilot on your team doesn’t mean weakening your security posture – existing policies like branch protections still apply in exactly the way you’d expect,” Dohmke explained.
The new tool works by booting a secure dev environment via GitHub Actions, cloning the repo, analyzing the codebase and pushing to a draft pull request. Users can observe session logs for greater visibility, validation and progress, with the Copilot agent promising to help across feature implementation, bug fixes, test extensions, refactoring and documentation improvements.
Dohmke also noted that users can give the coding agent access to broader context outside of GitHub by using Model Context Protocol (MCP).
The Copilot agent acts much like a human colleague in that it will tag you for review, where you can then leave a further comment asking it to make more changes, which it processes automatically.
Emphasizing the enterprise-grade security measures, GitHub noted: “The agent’s internet access is tightly limited to a trusted list of destinations that you can customize.” GitHub Actions workflows also need developer approval.
Copilot Enterprise and Copilot Pro+ will be the first account types to get access to GitHub’s new powerful agent, with each model request the agent makes costing one premium request from June 4, 2025.
GPT-4.1, GPT-4o, Claude 3.5 Sonnet, Claude 3.7 Sonnet and Gemini 2.5 Pro each account for one premium request, however more powerful and complex models have considerably higher multipliers. For example, one question using o1 costs 10 premium requests, and GPT-4.5 has a 50x multiplier. On the flip side, Gemini 2.0 Flash has a 0.25x multiplier, meaning four questions cost one premium request.
