JetBrains, the company behind coding tools like the IntelliJ IDE for Java and Kotlin (and, indeed, the Kotlin language itself), on Thursday launched Junie, a new AI coding agent. This agent, the company says, will be able to handle routine development tasks for when you want to create new applications — and understand the context of existing projects you may want to extend with new features.
Using the well-regarded SWE-Bench Verified benchmark of 500 common developer tasks, Junie is able to solve 53.6% of them on a single run. Not too long ago, that would have been the top score, but it’s worth noting that at this point, the top-performing models score more than 60%, with Weights & Biases “Programmer O1 crosscheck5” currently leading the pack with a score of 64.6%. JetBrains itself calls Junie’s score “promising.”
But even with a lower score, JetBrains’ service may have an advantage because of its tight integration with the rest of the JetBrains IDE. The company notes that even as Junie helps developers get their work done, the human is always in control, even when delegating tasks to the agent.
“AI-generated code can be just as flawed as developer-written code,” the company writes in the announcement. “Ultimately, Junie will not just speed up development — it is poised to raise the bar for code quality, too. By combining the power of JetBrains IDEs with LLMs, Junie can generate code, run inspections, write tests, and verify they have passed. “
It may be a bit before you can try that out yourself, though. The service is only available through an early-access program behind a waitlist. For now, it also only works on Linux and Mac, and in the IntelliJ IDEA Ultimate and PyCharm Professional IDEs, with WebStorm coming soon.
Topics
