GitHub launches multi-agent Copilot features for complex software workflows

From code completion to agentic workflows

GitHub is expanding its Copilot tool to move beyond simple code generation and developer assistance. The company announced a redesigned Copilot app and a new Projects capability during its annual GitHub Universe conference. These updates allow organizations to coordinate multiple AI agents to manage software development workflows from a single platform.

The previous version of the app primarily helped developers run agent sessions, review diffs, and create pull requests. The revamped version serves as a central workspace where developers can work on specifications, participate in design discussions, and analyze application telemetry. It connects with external tools to give agents a broader understanding of the work required rather than just individual coding tasks.

This shift allows developers to delegate broader tasks to cloud-based agents that execute work independently. These agents can then submit their proposed changes as pull requests for human review. This transition potentially changes the role of developers from writing and modifying code directly to defining tasks and validating the output of AI agents.

Coordinating work across repositories

The new Copilot Projects feature enables developers to assign broad goals that may span multiple repositories. Users can provide context from various sources, including design mockups, Slack conversations, and meeting notes. Once the context is provided, Copilot can distribute work across multiple agents running in parallel and track their progress.

This capability is specifically designed to help teams clear development backlogs by automating repetitive tasks. Examples include dependency updates, replacing deprecated software development kits (SDKs), and performing version upgrades for Java or .NET across multiple repositories. It is positioned as a tool for large-scale application modernization where manual updates are time-consuming.

However, running agents in parallel introduces coordination challenges for development teams. Because multiple agents work at once, they might edit the same code, rely on old API agreements, or make conflicting architectural choices. Developers will need to resolve these integration issues before any of the agent-generated changes can be merged into the main codebase.

Multi-agent reviews and new pricing

GitHub is also introducing a new Max tier for code reviews that utilizes multiple AI agents. Rather than relying on a single engine, Max employs several distinct review agents powered by different providers. An aggregation agent then consolidates these findings, resolves any disagreements, and checks for accuracy before presenting a combined review to the developer.

This multi-agent approach aims to improve the detection of defects and security issues in shared libraries and authentication logic. While it may catch issues a single model misses, it does not guarantee that all vulnerabilities will be detected. Developers will still need to validate critical findings and retain human approval for any high-risk code changes.

To manage the costs of these intensive processes, GitHub is shifting to an AI credit-based consumption model for enterprises. This model allows companies to pay based on actual usage rather than flat subscriptions. For example, GitHub estimates that Lite reviews consume an average of $0.25 in credits, while Max reviews cost $10.

Governance and security considerations

The ability for agents to independently execute tasks and modify repositories creates new requirements for enterprise security. Organizations will need to extend existing software development controls to these autonomous systems. Experts suggest that CIOs should restrict agent access to specific repositories and enforce branch protection policies to maintain control.

Monitoring agent activity becomes essential as companies scale their use of agentic development. Companies may need to integrate audit logs with security information and event management (SIEM) systems to monitor which agents are accessing code. This visibility is necessary to maintain accountability when multiple agents are making changes across a large organization.

Additionally, companies must manage the risks of poor task definitions which could lead to repeated executions and higher credit consumption. To mitigate this, CIOs can set credit limits for individual projects and assign human owners to monitor progress. These owners can then stop agent sessions if the AI begins to perform unnecessary rework or puts too much pressure on continuous integration pipelines.