What do you like best about Junie?
The core strength of Junie is its agnostic approach to intelligence. Instead of locking you into a single proprietary model, JetBrains gives you complete control via Bring Your Own Key (BYOK). I can tap into bleeding-edge commercial models (like Claude, GPT, or Gemini) or open-weight models (like Qwen or DeepSeek). With the open-weight models, users have access to much cheaper models.
Furthermore, the pricing model offers excellent ROI predictability. By pairing a standard AI Pro or AI Ultimate subscription with our team's internal BYOK endpoint rates, we significantly lowered token costs.
The agentic reasoning is top-tier: its Plan Mode prevents the typical "hallucination loop" seen in other tools. Instead of writing code blindly, Junie generates a multi-step engineering checklist (.junie/plans) that I can audit, edit, and commit before it modifies a single file. The Live Prompting capability is also a plus, I can steer the agent mid-task or course-correct it in real-time without starting the session over from scratch.
By passing the time-consuming boilerplate and configuration wiring off to Junie, a task that normally eats up half a day of a senior engineer's time was completed flawlessly in under 15 minutes.
Beyond the IDE panel, its git and platform integrations are remarkably mature. It handles asynchronous task delegation natively via GitHub Actions. I can spin up a massive codebase migration, push it to my repository, let Junie execute the multi-file refactoring headlessly in the cloud or pipeline, and review the resulting pull request.
In the desktop IDE, the visual interface blends natively with JetBrains AI Chat tool windows. However, the real star is the Junie CLI. The terminal interface functions as a highly intuitive Text User Interface (TUI). Executing a step-by-step session directly from the terminal prompt makes it feel like a cohesive extension of an advanced developer workflow.
On the support side, if you do hit a roadblock during onboarding, JetBrains has provided an exceptional support ecosystem. They host a dedicated, active Discord server where users can ask for help directly and get answer from the Junie engineering team. Review collected by and hosted on G2.com.
What do you dislike about Junie?
High Resource Overhead: Because Junie is built utilizing Java/Kotlin, its local footprint can become incredibly demanding. When running longer development or multi-step agent sessions, the memory and CPU usage spike significantly, occasionally dragging down system performance.
Restricted Subagent Flexibility: Advanced agent orchestration features, like subagents still feel overly restricted. They currently do not allow you to configure custom, user-defined models for specific sub-tasks, which limits the optimization of workflows.
Limited Open-Weight Selection: Junie deserves praise for allowing us to bring our own custom API keys and BYOK integrations. However, when it comes to the open-weight models that JetBrains natively cloud-hosts on its own infrastructure, the selection is mostly limited to families like Qwen and DeepSeek. I hope JetBrains broadens this first-party cloud-hosted catalog to offer a much wider variety of open-weight models. Review collected by and hosted on G2.com.