GitHub expands Copilot for JetBrains with GA custom agents, sub-agents, and plan agent

GitHub shipped a significant update for GitHub Copilot in JetBrains IDEs on March 11, 2026, and the release is more than a routine plugin refresh. The headline change is that custom agents, sub-agents, and plan agent are now generally available, giving JetBrains users a more structured way to split work across specialized agent roles directly inside the IDE.

The announcement also matters because GitHub is not just adding more agent surfaces; it is adding more control points around those agents. That is an important distinction for teams that want agent assistance to become part of daily engineering work without turning the IDE into an opaque black box.

What changed

GitHub says agent hooks are now in public preview. Developers can define a hooks.json file under .github/hooks/ and trigger custom commands around key session events such as userPromptSubmitted, preToolUse, postToolUse, and errorOccurred. That creates a practical bridge between Copilot agent sessions and repository-specific automation, policy checks, or observability logic.

The update also adds MCP auto-approve support at both the server and tool level. In day-to-day use, that should reduce manual approval churn for trusted tool calls while still letting teams keep boundaries around higher-risk actions. GitHub is effectively acknowledging that agent productivity depends on reducing repetitive confirmation loops without removing governance entirely.

Another notable addition is support for AGENTS.md and CLAUDE.md instruction files. GitHub says Copilot can automatically discover and load these instruction files into context, including nested files if the relevant setting is enabled. The JetBrains integration can also generate an initial AGENTS.md, which lowers the setup cost for teams that want project-level guidance to shape agent behavior from day one.

On top of that, GitHub moved auto model selection to general availability for JetBrains IDEs. Copilot can now choose a model based on current availability and performance conditions, reducing manual switching during active development.

Why it matters

This release shows GitHub pushing Copilot toward a fuller agent platform inside mainstream developer tooling. JetBrains users are getting not only more autonomous behavior, but also better mechanisms for instructions, hooks, and approval policy. For engineering teams, that combination is often what determines whether agent features stay as demos or become part of production workflows.

The strategic signal is clear: IDE integrations are moving beyond inline completion and chat toward multi-agent coordination, repository-aware instructions, and controlled tool execution. GitHub’s March 11 release puts JetBrains users directly into that shift.

Source