GitHub Copilot is being pushed beyond the familiar role of an AI pair programmer. With Agent Mode, the full version of Copilot Edits and the preview of Project Padawan, GitHub is outlining a future where AI coding tools do more of the surrounding development work while developers keep review and approval authority.
The direction is clear: Copilot is being designed to act less like a passive suggestion box and more like a development partner that can notice problems, propose actions and help move work through a project.
Agent Mode gives Copilot more room to act
Agent Mode is the clearest sign of GitHub’s push toward more autonomous AI coding. According to GitHub CEO Thomas Dohmke, Copilot’s new Agent Mode can spot and fix problems on its own.
That does not mean the tool simply takes over the developer’s machine. The system can suggest terminal commands, but it runs them only after getting a thumbs-up from the developer. That approval step matters because it keeps the developer involved when an action moves from suggestion to execution.
Agent Mode is also designed to understand work that is connected to the original request. One of its key features is the ability to identify and handle additional tasks that were not explicitly mentioned but are necessary to complete the main objective.
In practical terms, that changes the shape of an AI coding session. A developer may describe the target outcome, while the assistant looks for related steps needed to get there. The source makes clear that this feature is currently available to VS Code Insiders users.
Copilot Edits turns natural language into multi-file changes
GitHub has also launched the full version of Copilot Edits. This feature lets developers modify multiple files at once using everyday language, including voice commands.
That expands the way developers can interact with a codebase. Instead of moving file by file and making each change manually, a developer can describe the change they want and let the tool generate suggested edits across the affected files.
The system uses two models working together. One model generates the suggested changes, while another handles the editor integration. That split suggests GitHub is treating code generation and the act of placing those changes into the development environment as separate responsibilities.
Developers can choose their preferred AI model. The available choices named in the source are GPT-4o, Claude 3.5 Sonnet and Gemini 2.0 Flash.
GitHub says the feature is built to keep developers in control. They can review changes across files and choose what works. If something goes wrong, they can test the code and roll back to what worked before.
Project Padawan points to tested pull requests from AI
Project Padawan is GitHub’s early look at what more fully autonomous software development might look like. The system is set to launch later this year.
The source describes a tool that will be able to handle issues and create tested pull requests on its own. That is a larger step than generating snippets or helping with local edits. It moves Copilot closer to the workflow where development tasks are tracked, implemented, tested and prepared for review.
Project Padawan will work directly within the GitHub interface. Developers will be able to submit issues directly to Copilot, giving the system a defined task inside the same environment where teams already manage code and pull requests.
The implied workflow is simple: an issue becomes the input, and a tested pull request becomes the output. The human team still has to decide what to accept, but the AI takes on more of the work required to get a proposed solution in front of reviewers.
Why this matters for software teams
The three Copilot updates point in the same direction. GitHub is trying to make software development more autonomous without removing developers from the approval loop.
Each tool covers a different part of the coding process:
- Agent Mode helps Copilot notice problems, suggest terminal commands and handle related tasks needed to complete an objective.
- Copilot Edits lets developers request changes across multiple files using everyday language or voice commands.
- Project Padawan aims to handle issues and create tested pull requests within GitHub.
That combination matters because software work is rarely just writing one isolated block of code. It often includes understanding the task, finding related changes, editing across files, testing the result and preparing work for review.
GitHub is positioning Copilot to cover more of that chain. The company also presents this as a possible way to save companies a lot of time and money, though the source frames that as potential rather than a guaranteed outcome.
Control remains part of the pitch
The most important boundary in these announcements is developer control. Agent Mode asks for approval before running terminal commands. Copilot Edits lets developers review changes, test them and roll back if needed.
That balance is central to the current version of autonomous AI coding. GitHub is not only showing tools that can act more independently. It is also emphasizing checkpoints where developers can inspect, accept or undo the work.
For now, GitHub Copilot’s direction is not just about faster code generation. It is about turning AI into a more active participant in the development workflow, from spotting follow-up tasks to preparing pull requests. Agent Mode, Copilot Edits and Project Padawan together show how quickly AI coding assistants are moving from suggestions toward project-level execution.