JetBrains Junie internal beta experience, can it challenge Cursor?

What is Junie?

JetBrains launched a new AI Agent product, Junie, on January 23, 2025, which is currently in the internal testing phase. This product can be similar to Cursor's Agent mode, which can automatically execute code or commands based on user intent.

According to JetBrains, Junie can:

1. 1. Deep integration with JetBrains IDE, making full use of project context information in IDE
2. 2. Independently complete complex tasks such as CRUD operations, including analyzing project structure, checking dependencies, creating files, etc.
3. 3. Run tests and code inspections to ensure the quality of the generated code
4. 4. Support mainstream languages ​​such as Python, Java and Kotlin

What can you do?

As JetBrains' official AI Agent, although it is a plug-in, Junie's biggest advantage lies in its deep integration with IDE:

• • Can directly access and understand the project structure
• • Ability to run and validate tests
• • Code inspection can be performed
• • Adapt to the workflow of JetBrains IDE. For example, after executing the user's intention, Junie will automatically build, run and test. During this automatic build process, you can see the entire build process in the IDE.

How to do it

The Agent of this plugin can be divided into two steps:

1. 1. Plan: After entering the problem, a step list will be displayed on the left.
   
2. 2. Act: Act is to execute the steps generated by the plan on the left. Each step includes the call of multiple tools, such as: searching for code, calling large models, applying code, executing commands, building projects, etc.
   

After executing the steps, you can also see the relevant code diff view, but you can't select the fragment to accept. I can only choose Done to accept all, or Decline to reject all. This experience is too uncomfortable.

The content output in the Agent dialog box cannot be output in Chinese, but writing code or comments is not affected.

For the business code in the project, if it is not described clearly, it is also possible to make incorrect judgments after searching.

When exporting content, there are only a few preset options (there is nothing else to configure):

1. 1. Select files, you can select multiple files
2. 2. Check Brave Mode, this mode will automatically execute the command
3. 3. If you choose Project guidelines, a hidden file will be created, and then the Agent analysis project will be executed to improve the file. Now I feel that the analysis is not very good. Positioning is similar to Cursor .cursorrules document