Air in IDEs – Part 1 | Create a Session and Track Your Agents' Work
In a Nutshell
The video demonstrates JetBrains AI's session-based workflow in IntelliJ IDEA, where multiple AI agents can work simultaneously on tasks like adding search, filtering, and sorting to a reading list app. It shows how to create sessions with different agents and models, monitor their progress through chat or terminal interfaces, review file changes, and track costs without executing code changes automatically. The core takeaway is that JetBrains AI enables parallel agent workflows with built-in session management, change tracking, and cost monitoring for safe, controlled AI-assisted development.
These notes were generated by AI and may contain inaccuracies.
JetBrains AI and IDs bring agent-driven workflows into JetBrains development environments. This series explores working with artificial intelligence in practice, from setting up the first session to reviewing, testing, and delivering code. The focus begins with setting up sessions and tracking what happens. A reading tracking application is ready in IntelliJ IDEA for adding improvements with a focus on working with artificial intelligence.
The project is a simple application containing an index of books and a personal reading list. Users can track progress, rate books, and leave notes. Finding items in a long list requires significant scrolling. The goal is to add search and filtering with sorting and numbering using the current application structure.
For version 2026.2, install AI from the JetBrains store. For the 2026.3 EAP release, AI is already integrated into the development environment. The version used is 2026.3 EAP, accessible in the toolbar of the window.
Before making changes, a question about the project is posed. Pressing on a new session allows choosing an agent, which is Juny in this case. AI automatically detects proxies in the system, and users can install other proxies or add custom proxies. Artificial intelligence is not tied to just one agent. Different agents can be used for different tasks, or multiple sessions can be created with the same agent.
In the session, Juny is selected, then a model is chosen. Juny Light is selected because it is free. It is free because it uses a lightweight language model provided by the JetBrains AI platform. Juny Light is the perfect choice for experiencing artificial intelligence without subscriptions or extra keys.
In the first session, the prompt reads: "Explain how the book differs from the reading entry and where each reader's progress, ratings, and comments are stored. Do not change any files." The session is sent, creating the first session on artificial intelligence that can be returned to from the session list.
Another way to call up a new session is to press the Control key twice. From here, either an existing session can be accessed or a new one created. A new session is created with CodeX for diversification. A concrete task is assigned: improved search functionality on the reading list page. A prompt is pasted requesting title and author search, filtering, sorting, and page numbering using the application's current structure. The agent is also asked to keep each reader's results separate, maintain filters when navigating between pages, and add quizzes. The agent should not carry out any lifting or approval process for anything.
The session now awaits input. Activating a small option triggers a sound every time the agent needs input, unless focusing on the chat window for that session. This is extremely useful when working on other tasks outside of the development environment while agents are working, or while preparing dinner.
Some prefer the terminal user interface and others prefer the chat view, and both are provided. The current session uses the chat view where the conversation and agent activity can be followed. If a proxy terminal interface is preferred, "Ear" also supports compatible proxy terminal sessions. Moving to Agent Management allows moving "Johnny" to the terminal view, so future "Johnny" sessions will open in a terminal user interface.
This is a separate example dealing with how to run a project. Behind the scenes, several other sessions were set up for demonstration purposes. During proxy sessions, tasks requiring attention can be seen. Here, the agent awaits a decision regarding the tags. The books should match all the specified tags. This answer is sent to let the agent continue working. The unread update is different and new, but the agent is not necessarily out of business. Opening the session allows keeping up with updates without losing track of the feature's progress.
While the feature is being built, a separate question about the existing code is posed. Another session of the service being looked at is begun. Pressing Command + 2 on Mac OS or Alt + 2 on Windows opens the proxy sessions. Voice input is used this time. Clicking on the microphone, speaking the request, and confirming the recording allows the prompt: "Explain how the reading list service keeps each reader's books separate. Keep it brief and don't change any files." Once the text appears, it can be reviewed and corrected before sending.
The agent previously asked to improve the search on the Reading List page has completed its work. Looking beyond its summary, the session displays the files that have been changed, and differences can be opened directly from here. For this feature, both sides are examined: how the back end applies the filters, and how the page keeps them selected during navigation. It looks good. Only the updated readme file will be uploaded for the time being.
Going to Options and selecting "Show outgoing deposits" shows changes that have been deposited locally but not yet paid out. This way, what is ready to pay can be tracked alongside the agent sessions being worked on.
Now that all parallel sessions have been opened, the interface looks crowded. Relevant sessions can be grouped into a folder so they are easier to find later. Colored stickers can also be added for better differentiation. Session costs can be seen while working. These estimates help compare spending across different agents and model combinations, gain a better understanding of what future tasks might cost, and stay within budget.
The result is checked by running the tests in IntelliJ IDEA. In the app, completed fiction books tagged for the book club that have received four stars or more can be found without scrolling through the entire list. The control remains selected, and results can be sorted or navigated between pages without losing filters. Deleting returns to the full list.
These are the basics related to artificial intelligence and software identifiers. Choose an agent, give them useful context, monitor their progress, and check the outcome. There is a lot to explore, with more advanced features and workflows covered in the following video.
To try it, artificial intelligence can be tried with Junior Light for free. No subscription is required. Simply log in with a JetBrains account. The description contains the current offer and terms.
Keep JetBrains in your library
Save the videos and channels worth coming back to, and find them again in one place.
