GitHub unveils Copilot Fusion and multi-agent Projects at Universe 2026

GitHub used its Universe 2026 stage in San Francisco to introduce Copilot Fusion, a new system for choosing and coordinating AI models inside its coding assistant. Alongside it came multi-agent Copilot Projects and a set of infrastructure upgrades built for what the company calls agentic development.

The message from the October 28-29 event was hard to miss. GitHub no longer wants Copilot to be the clever autocomplete in your editor. It wants Copilot to run the whole crew.

What GitHub actually announced

Copilot Fusion is the headline act. GitHub describes it as a unified agent service, and its core job is model selection and orchestration.

Instead of sending every request to one AI model, Fusion decides which models should handle which parts of a task, then coordinates how they work together.

The second piece is multi-agent Copilot Projects. Projects lets several AI agents collaborate on development work rather than leaving one model to grind through everything on its own.

The third piece is infrastructure. GitHub rolled out upgrades aimed at supporting agentic development, and Microsoft Foundry sits at the center of that story.

Foundry, which is connected with Copilot, handles agent hosting and governance.

The road from HydraFusion to Fusion

Copilot Fusion did not appear out of nowhere. It builds on Project HydraFusion, which GitHub released as a research preview in September 2026.

HydraFusion was designed to orchestrate multiple models at runtime, meaning it could run several of them at the same moment. The preview was reported to deliver better performance at a lower cost than single-model tools such as Claude Code.

Fusion takes that research concept and turns it into a product. One month separates the preview from the launch.

GitHub also pointed to the scale underneath all of this. The platform reportedly recorded 7.4 billion commits in September 2026.

Why the shift to agents matters

For most of its life, Copilot has been a suggestion engine. The Universe announcements describe something closer to delegation. A developer hands off a larger task, and a coordinated set of agents attempts to manage it with less hand-holding.

GitHub is trying to weave AI deeper into the full software development lifecycle, not just the moment when someone is staring at a blinking cursor.

The multi-model approach also says something about where GitHub thinks the AI race is heading. Rather than staking everything on one model provider, Fusion is built around picking the right model for each job.

What this means for developers and the competition

On the competitive front, the explicit comparison to Claude Code sets up a direct contrast in philosophy. Single-model tools bet on depth from one powerful system. GitHub is betting on coordination across many.

GitHub also brings a distribution advantage that is hard to ignore. Its platform is already where an enormous share of code lives, and the September commit figure underscores how much developer activity flows through it every day.

Bolting multi-agent capabilities directly onto that platform, with Microsoft Foundry handling the hosting underneath, gives GitHub a tightly integrated stack.