Skip to content

Course · Copilot · Beyond VS Code

The same work in JetBrains

Picture the teammate who never opened VS Code - they have lived in a JetBrains IDE their whole career and are not going to switch editors to use Copilot. Good news: they do not have to. Copilot’s agent runs in JetBrains too, and the work from this course mostly transfers as-is. “Mostly” is the operative word, and the point of this lesson is to make the exceptions precise rather than wave at them.

Agent mode and MCP (the protocol for connecting external tools and data to the agent) are available in JetBrains, but support varies by feature and plugin version. GitHub’s comparison (as of October 2026) marks MCP as supported while custom instructions, custom agents, subagents, prompt files, and Agent Skills remain preview in JetBrains. Check the Copilot customization support matrix before making a feature part of a team workflow.

The thing that doesn’t change: your rules

Section titled “The thing that doesn’t change: your rules”

Start with the reassurance, because it is the biggest one. In the Rules chapter you wrote convention files - an AGENTS.md, and the CLAUDE.md you may have carried over from another tool - so Copilot’s first draft would already match how each repo does things. JetBrains auto-discovers and reads those same files - with the caveat the matrix carries: rules support in JetBrains is preview, and the matrix’s JetBrains rows name AGENTS.md support without mentioning CLAUDE.md, so treat that half as unverified. You re-author nothing and keep no second copy in sync; the AGENTS.md at the root of each repo is discovered on opening a project - worth the one-time check the tip below shows you how to make.

The conventions are a property of the repo, not the editor - so the careful work teaching the agent your team’s idioms is exactly as valid in JetBrains as it was in VS Code. The single most expensive context-engineering work in the course travels.

Open the shared library with its consumers - shared-lib, the course’s illustrative internal library, imported by about a dozen services - or any repo you can safely open in JetBrains. Agent mode and MCP are the stable parts of the comparison. The customizations built earlier in the course land differently:

  • Custom instructions, custom agents, subagents, prompt files, and skills are preview or version-dependent in JetBrains. Usable, worth trying, but test the exact plugin version before a team’s workflow depends on them.
  • Agent hooks (deterministic gates that run automatically at agent lifecycle points) are not currently supported in JetBrains per the matrix - not preview, absent. Do not build on them there.
  • Plan-first work. You can still apply the planning discipline manually - ask for the approach in chat, read it, approve, then build - but do not assume the VS Code plan surface with its approve-and-continue action is identical in JetBrains.
  • Inline Agent mode (driving the agent from inline chat rather than the panel) is preview or version-dependent - announced as a preview in GitHub’s April 2026 changelog rather than any matrix row.

The careful posture itself transfers completely: review diffs, keep approvals tight, plan before editing. What does not transfer automatically is the exact customization surface. Keep preview features optional until your team’s supported IDE and plugin versions are confirmed.

The practical path through the two repos in JetBrains looks similar at the task level: open the project, verify which instruction files your plugin discovers, and keep a human-readable plan and a real diff review for anything consequential. If your flow leans on custom agents, subagents, prompt files, or skills, verify support first and treat those pieces as optional until your versions mature.

One more status note: reaching the Copilot CLI’s agent from inside JetBrains - the “Copilot harness” - moved to generally available in August 2026 (per GitHub’s changelog). If you see older material calling it a preview, that has since changed; still verify the specific delegate-to-CLI surface against your current plugin, since the harness and any particular in-IDE affordance around it do not always move together.

The honest summary for the JetBrains teammate: you lose almost nothing that matters, your AGENTS.md comes along (verify the CLAUDE.md half), and the thinner spots - rules, skills, and the other customizations - are preview-grade, so verify before relying on them. The judgment you built - matching autonomy to stakes - is editor-independent; it was never a VS Code feature in the first place.

There is one more surface, and it is the smallest: the same assistant, reachable from a terminal. Next: the terminal surface - Copilot CLI.