GitHub has launched isolated subagents in public preview for JetBrains IDEs, Eclipse, and Xcode, introducing a new way for developers to delegate focused tasks to autonomous AI assistants without disrupting their primary workflow. The feature brings multi-agent capabilities to three major development environments outside VS Code.
What Subagents Actually Do#
Subagents let developers spin off specialized tasks—like researching documentation, reviewing code, generating API specs, or gathering project context—to dedicated Copilot instances that work autonomously. Each subagent runs with its own isolated context, preventing lengthy research or analysis sessions from crowding out the main conversation history.
Once invoked, a subagent proceeds without further user interaction, making it ideal for background tasks that would otherwise interrupt a developer’s flow. When the subagent completes its work, results appear back in the main conversation ready for follow-up questions or next steps.
Two Ways to Invoke#
Copilot can automatically delegate tasks to subagents based on the request type and available custom agent configurations. The system evaluates task descriptions and context to determine when delegation makes sense.
Developers can also explicitly invoke specific subagents by referencing them directly in commands—for example, asking the “code-reviewer subagent” to examine recent changes or having the “documentation subagent” generate API docs for a module.
Custom Agents Required#
The feature builds on GitHub Copilot’s custom agents infrastructure. Developers need custom agents configured in their environment before subagents become available. Organizations can define specialized agents with curated instructions and allowed tools, which then become available for subagent delegation.
Keeping Context Clean#
The core value proposition centers on context management. AI assistants become less effective when conversation history grows cluttered with tangential research and background tasks. By isolating subagent work into dedicated contexts, the main conversation stays focused on the primary development task.
This approach mirrors how developers naturally work—delegating specific research or review tasks to colleagues while maintaining focus on their own work. The difference is the “colleague” is now an autonomous AI instance operating in parallel.
GitHub is collecting feedback through in-product controls and dedicated discussion repositories for each supported IDE platform.