5 min Applications

JetBrains details deeper mechanics of Junie coding agent

JetBrains details deeper mechanics of Junie coding agent

Coding agents are coming and software application developers are variously concerned, perniciously perturbed and perplexed, but perhaps also rather excited to see how the next wave of engineering accelerators plays out on (or more likely at an abstracted level above) the command line. Product specialist at project management and workflow enhancement tool company JetBrains, Anastasia Krivosheeva thinks that AI coding agents have the potential to take over more repetitive tasks, generating code faster. But what matters most now?

The truth here may come down to the fact that speed is good, but when we’re talking about using a turbo-charged coding agent (or indeed an in-IDE AI assistant, or a debugging bot of some kind), back in real-world scenarios, speed should be considered secondary to accuracy, intelligence and trustworthiness.

Fast good, smart better

“Fast agents will reduce the time it takes you to complete a given task, but smart ones will tangibly enhance the end-to-end development experience, minimising errors, bugs and context switching and helping you achieve high-quality code,” blogged Krivosheeva.

Alluding to the arrival of JetBrain’s own coding agent Junie, the product development team at JetBrains insist that primary attributes for any tool in this space are transparency of process, the ability to draw upon structured task planning and the power to make use of comprehensive progress tracking.

How does Junie work?

Junie is officially described as an AI coding agent that is primarily tasked to “autonomously plan and execute” multistep actions in complex processes based on an initial prompt from the developer themself. Junie can introduce large-scale edits to a software engineering project when instructed and also run tests or terminal commands. It does this while reporting to the software engineering team on the progress of the task completion.

Built to navigate project files, run code, execute terminal commands and modify a file system autonomously, Junie (which is only available in online mode at the moment with no public plans for an offline version) relies on the core tools of the JetBrains IDE, including source code and project structure navigation, search everywhere and code inspections. Junie sends requests to a variety of different configurations of Large Language Models (LLMs), including third-party providers such as Anthropic and OpenAI.

“Our users value a detailed execution plan for approval, an in-depth actions log with reasoning for each action taken and extensive collaboration options, including autonomous partner-like interaction,” said Krivosheeva. In survey data from JetBrains, some 83% of users (in management roles) mentioned that Junie increases their team’s productivity and 76% are satisfied or very satisfied with Junie.

Also this year, JetBrains has explained how and why it is introducing updates to Junie.

As recently reported on Techzine, Junie will now will be 30 percent faster and will support Model Context Protocol (MCP) and remote development.

“With Junie now fully integrated with GitHub, the power of async development (see below) is at your fingertips. Delegate multiple tasks simultaneously, make quick-fixes without opening an IDE, and collaborate with your team on the same project straight from GitHub,” said Krivosheeva.

For completeness here, let’s note that asynchronous (async) development programming is a technique, methodology and code construct approach designed to enable software applications to execute multiple operations in parallel, concurrently and at the same time. This means that one operation does not necessarily need to wait for the next (or any other operation) to finish before starting. Asynchronous development stands in contrast (obviously) to synchronous programming, where operations, tasks, calls and other actions are executed sequentially, which is more logical for some systems, but has the potenytial to cause performance bottlenecks.

“Junie isn’t just another coding agent – it’s built for how real developers work,” said Kirill Skrygan, CEO, JetBrains, speaking directly to Techzine this week. “While most agents chase speed, Junie focuses on what actually matters: reducing errors, cutting context switching and helping developers write higher-quality code. It’s a smart, reliable collaborator that fits into developers’ IDEs and adapts to their workflow – not to take over, but to support the way they build. From automating the repetitive to running complex tasks, Junie enhances the developer experience. In real-world software, speed is nice – but intelligence, trust and code quality matter more.”

Key JetBrains Junie functionalities

  • Developers now embracing Junie are promised the ability to save time on quick fixes and changes that don’t require them to open the IDE and code directly in their project. Instead, they can localise multiple pages from one project without having to stop working on another.
  • Software engineers are also now given the ability to run multiple tasks at the same time across different projects i.e. they can fix a bug in their work repository, ask Junie to implement a new feature in a pet project and then review and adjust them together.
  • Developers can also explore new ways of coding as a team: assign a task to Junie, ask a teammate to review it and then finalise the project together as a team.

Junie on GitHub is currently available for JVM and PHP only.

“Our aim is to make Junie available to as many teams as possible, which is why we offer grants for open source projects for Junie on GitHub,” said Krivosheeva. “Junie has already proven its effectiveness when it comes to understanding complex codebases in different programming languages, starting new projects, generating high-quality code, and writing functional tests.”

Junie is currently available for MacOS and Linux only.