Junie is JetBrains’ native agentic coding assistant, built directly into IntelliJ IDEA, PyCharm, WebStorm, GoLand, and the rest of the JetBrains IDE family. Unlike third-party plugins that bolt onto an editor’s extension API, Junie has access to the full JetBrains Platform — the same deep language indexing, refactoring engine, and test runner infrastructure that powers the IDEs themselves. The Junie CLI, released in general availability in March 2026, extends this capability to terminal environments and CI/CD pipelines, allowing the same agent behavior that runs inside the IDE to execute in automated workflows without a graphical session.
Key capabilities
Task Blueprint planning — Before writing or modifying any code, Junie produces a Task Blueprint: a structured plan that lists each file to be changed, the rationale for each change, and the order of operations. The developer reviews and approves the Blueprint before execution begins, making the agent’s intent explicit and reviewable rather than implicit.
Test-driven iteration loop — After implementing changes according to the Blueprint, Junie automatically runs the project’s existing test suite, reads the output, diagnoses failures, and iterates on the implementation until all tests pass or a human decision point is reached. This closes the write-test-fix loop that consumes significant developer time on non-trivial changes.
MCP Agent Skills — Junie supports Model Context Protocol integrations called Agent Skills, which allow external tools, APIs, and data sources to be exposed to the agent as callable capabilities. Teams can build Skills that connect Junie to internal services, documentation systems, or monitoring platforms, extending its knowledge beyond the local codebase.
Junie CLI for CI/CD — The CLI enables Junie to be invoked from a shell, a GitHub Actions workflow, or any other automation context. This means the same planning and execution behavior available interactively in the IDE can be triggered as part of a pipeline, enabling agentic code generation and patch application without human presence in the loop.
Autonomy level
Junie operates at autonomy level 4. It independently produces a full Task Blueprint, executes multi-file changes, runs tests, and iterates on failures across multiple rounds without per-step prompting. Human approval is required at the Blueprint review stage, and Junie surfaces a decision point when it cannot resolve a test failure autonomously. The CLI mode can reduce human checkpoints further when used in fully automated pipelines, making Junie one of the more autonomous options among IDE-native agents.
Strengths
- Deepest JetBrains IDE integration of any agent: accesses the full language index, refactoring engine, and test runner natively
- Task Blueprint planning makes the agent’s reasoning transparent and reviewable before any code changes land
- Automatic test execution and iteration loop reduces the back-and-forth of catching regressions manually
- Junie CLI enables the same agent behavior in CI/CD pipelines and terminal sessions without the IDE
- MCP Agent Skills allow connection to internal tools and external services through a standard protocol
- Supports Claude, GPT-4o, and Gemini, giving teams flexibility in inference backend
Limitations
- JetBrains-only: there is no VS Code, Neovim, or other editor support, which excludes a large portion of the developer population
- Requires a paid JetBrains AI Assistant subscription on top of any existing JetBrains tooling costs
- Proprietary with no open-source components; the agent planning logic and Skills runtime are not auditable
- Tightly bound to the JetBrains runtime and platform, which can make behavior harder to predict when the IDE itself has platform-specific quirks
- No browser or web-browsing capability, limiting the agent’s ability to fetch documentation or reference external resources autonomously