GitHub launches Copilot Fusion to coordinate multiple AI agents

From autocomplete to delegation

During the Universe 2026 event in San Francisco, GitHub announced a major shift in how its coding assistant functions. The company is moving away from providing simple suggestions for a blinking cursor toward a system of delegation. Instead of just helping a human write a line of code, the new tools aim to manage larger development tasks with less manual oversight.

The core of this transition lies in the introduction of Copilot Fusion and multi-agent Copilot Projects. These features are designed to let developers hand off complex work to a coordinated set of AI agents. This approach seeks to weave artificial intelligence deeper into the entire software development lifecycle rather than just the moment of typing.

This strategy reflects a change in how GitHub views the role of AI in professional software engineering. By focusing on orchestration, the platform wants to enable a workflow where the AI can run the 'whole crew.' This moves the technology from a passive assistant to an active participant in the development process.

How Copilot Fusion works

Copilot Fusion serves as a unified agent service that handles model selection and orchestration. Rather than sending every single request to one specific AI model, Fusion determines which models are best suited for different parts of a task. It then coordinates how those various models work together to complete the objective.

This system is built upon the research preview known as Project HydraFusion, which was released in September 2026. That research demonstrated that orchestrating multiple models at runtime could provide better performance than single-model tools. It also showed that this multi-model approach could achieve these results at a lower cost than competitors like Claude Code.

By turning this research into a product, GitHub is betting on the power of coordination. Fusion allows the platform to pick the right tool for the right job dynamically. This flexibility ensures that the system can leverage the specific strengths of different models while maintaining a cohesive output for the developer.

Multi-agent projects and infrastructure

Alongside Fusion, GitHub introduced multi-agent Copilot Projects to facilitate collaboration between different AI entities. In this setup, several agents work together on a single project rather than leaving one model to grind through every single step alone. This collaborative environment is intended to handle more complex, multi-faceted development work more efficiently.

To support these capabilities, GitHub rolled out significant infrastructure upgrades specifically for agentic development. At the center of this infrastructure is Microsoft Foundry, which provides the necessary hosting and governance for the agents. This ensures that the agents operate within a structured environment that can manage multiple tasks simultaneously.

The integration of Microsoft Foundry with Copilot creates a tightly integrated stack for developers. Because GitHub already hosts a massive amount of code, bolting these multi-agent capabilities directly onto the platform provides a significant distribution advantage. It creates a seamless experience where the infrastructure and the AI tools are designed to work in tandem.

The shift in AI strategy

The move toward multi-agent systems suggests a specific philosophy regarding the future of the AI race. Rather than staking everything on the depth of a single powerful model, GitHub is prioritizing the coordination of many models. This approach acknowledges that different models may excel at different types of coding tasks or logic problems.

This contrasts directly with single-model tools that bet on the sheer power of one system to do everything. By choosing a multi-model approach, GitHub aims to create a more robust and versatile system. It allows the platform to be less dependent on any single provider and more focused on the best possible outcome for the user.

The scale of this rollout is underscored by the fact that GitHub recorded 7.4 billion commits in September 2026 alone. By introducing these agentic features to such a large user base, GitHub is positioning itself to lead the transition toward autonomous development. The goal is to provide a system that can manage the heavy lifting of software creation while the human developer provides the direction.