GitHub rebuilds Copilot around agents

At Universe, its annual conference, held at the end of October, GitHub presented a new phase for Copilot. The tool, which started out suggesting lines of code, is now an app built for what GitHub calls agentic software delivery: developers set the work and follow agents that carry out parts of it.

What was announced

• Copilot app: an application that connects agents and tools across projects, so you can direct the work without making every handoff.
• Copilot projects: one place to track the agents’ progress, make decisions and set the next steps, already available in the app.
• Copilot Fusion: evaluates each task and chooses how to execute it, with one or more AI models.
• Copilot code review Max: a deeper review, built to surface the bugs that need fixing before they ship.
• Automations: AI workflows that run in the cloud inside a repository, triggered by hand, by an event or on a schedule, and that propose improvements for the team to approve.

There is also a commercial change: GitHub Enterprise customers can now use every Copilot feature without a separate subscription and pay for the AI credits they use. Existing Copilot Business and Copilot Enterprise customers can keep their current plans.

What it changes for companies that buy software

GitHub’s message is that developers still set the direction, make the decisions and stay accountable for what ships. In practice, the bottleneck moves from writing code to deciding what to build, reviewing what the agents deliver and keeping it all secure.

For companies that build or buy software, three points deserve attention:

• Review stays human: more code written by agents calls for clear acceptance criteria, automated tests and someone accountable for approval.
• Cost now varies with use: with billing by AI credits, it pays to set limits and track usage per project.
• Agents need governance: permissions, logs and approvals, the same care we described in our post on managing AI agents.