GitHub debuts Copilot agent to boost coding with secure oversight

GitHub has announced that its Copilot software now includes an asynchronous coding agent, directly integrated into GitHub and accessible from Visual Studio Code (VS Code).

The new Copilot coding agent operates within GitHub's native control layer and is powered by GitHub Actions, enabling developers to delegate both routine and specialised coding tasks. It is designed to require human approval before any continuous integration or continuous deployment (CI/CD) workflows proceed, ensuring that developers retain oversight of code changes.

The coding agent, previously referred to as Project Padawan, is intended to help developers work more quickly and at greater scale, while supporting secure code commitments. The Copilot coding agent is now available to all Copilot Enterprise and Copilot Pro+ users, with organisations including CommBank, Carvana, and EY among those already recognising its usage.

GitHub Chief Executive Officer Thomas Dohmke said, "GitHub is where the world's developers work on their projects. Now, it's becoming the place where they collaborate with agents in a configurable, steerable, and verifiable way. It's vital that organisations and developers are ready to embrace these agents without compromising their security posture."

He added, "Built around an integrated, secure, and fully customisable development environment powered by GitHub Actions, the Copilot coding agent is the most enterprise-ready of its kind—amplifying human developers with trust by design. And these protections aren't just for us: as the new home of AI agents, we're making the same primitives available to partners to ensure an open ecosystem for agentic peer programming."

The Copilot coding agent integrates with the software development life cycle. Developers can activate the agent by assigning a GitHub issue or initiating a session through Copilot Chat in VS Code. Commits generated by the agent are pushed to a draft pull request and can be tracked through session logs. Developers can review the agent's work, provide feedback, and request further iterations via regular pull request reviews.

The agent is designed to maintain security through features such as branch protections and managed internet access. Additionally, any pull requests created by the agent require manual approval before triggering any CI/CD activities, establishing an additional layer of control over build and deployment processes.

With the Model Context Protocol (MCP), developers are able to provide the agent with access to external data and additional capabilities. Configuration of MCP servers is available within repository settings, offering flexibility in how the agent interacts with software and data.

The Copilot coding agent's computing environment is underpinned by GitHub Actions, leveraging a platform with more than 25,000 actions available in the GitHub Marketplace. Every weekday, over 40 million tasks are handled by GitHub-hosted and self-hosted runners. The agent thus operates within an established infrastructure that supports a broad range of development workflows at both open source and large enterprise scale.

During private preview testing, the agent demonstrated effectiveness at completing low-to-medium complexity tasks in established codebases. These tasks include adding features, resolving bugs, extending tests, refactoring code, and improving documentation.

Rodrigo Castillo, Chief Technology Officer at CommBank, commented, "We believe agentic AI is the next phase of evolution for software engineering. Together with GitHub, we have tightened DevSecOps cycles, while using existing security controls without modification. We're also automating common tasks, such as tech debt management, giving our engineering teams more time for creative flow: designing better products and experiences for our customers, and better platforms for our other engineers."

Alex Devkar, Senior Vice President, Engineering and Analytics at Carvana, said, "The GitHub Copilot coding agent fits into our existing workflow and converts specifications to production code in minutes. This increases our velocity and enables our team to channel their energy toward higher-level creative work."

James Zabinski, DevEx Lead at EY, added, "The Copilot coding agent is opening up doors for human developers to have their own agent-driven team, all working in parallel to amplify their work. We're now able to assign tasks that would typically detract from deeper, more complex work—allowing developers to focus on high-value coding tasks."

Copilot Enterprise and Copilot Pro+ users can access the agent via consumption of Copilot premium requests and GitHub Actions minutes, both of which are included as part of their subscription entitlements.

Kate Holterhoff, senior analyst at RedMonk, observed, "With its autonomous coding agent, GitHub is looking to shift Copilot from an in-editor assistant to a genuine collaborator in the development process. This evolution aims to enable teams to delegate implementation tasks and thereby achieve a more efficient allocation of developer resources across the software lifecycle."

GitHub has also announced new artificial intelligence-related features intended to increase flexibility for software developers. Next month, GitHub will begin open sourcing the Copilot Chat extension for VS Code under the MIT licence, allowing developers to inspect and extend the AI's functionality. Over time, key AI features will be integrated directly into the VS Code core to further support developer control and transparency.

All GitHub users now have access to the Models tab within repositories, where they can build, test, and manage AI features. This update includes support for prompt management, lightweight evaluations, and enterprise controls for secure and governed AI usage on the platform.

GitHub Models has expanded to include support for Grok 3 from xAI, broadening the choice of models available to developers seeking different AI capabilities in their projects.

Agent mode functionality is being extended with availability in JetBrains, Eclipse, and Xcode integrated development environments, allowing Copilot users to work with agents in a wider range of programming tools.