Personal experience with JetBrains Junie™: day 2
Michelangelo van Dam's review of using Junie for the first time - Part 2
Michelangelo van Dam has created a two part blog post series on their experience with Junie, the agentic AI offering from JetBrains after watching this video from Trisha Gee. If you missed it, you can jump back to part 1 before reading this.
The goal of this session was to see if Michelangelo could use additional prompts to ensure that the code generated by Junie was tested and worked locally before getting Junie to create a GitHub action and publish the application to an Azure app service. I've never thought of using Junie for deployment too so let's see how he got on.
First up, Michelangelo noticed a few things needed addressing in the code that Junie had created previously so he set about sorting that out. This is an important step in my opinion. Junie (or any agentic coding agent for that matter) isn't here to replace you, it's here to complement your work. You're still the smart developer so it's imperative that you review the code and tweak it accordingly.
Following that Michelangelo sets up his database and sets to the first experiment which was to pair with Junie to make sure no secrets were exposed with the code. I can tell that Michelangelo put thoughts into the prompts he used here again because they're very explicit and leave nothing to the LLM-imagination. Michelangelo also followed up himself to make sure that .env was no longer part of the codebase using Git. Next up, it's ensuring that a specific directory is listed in the .gitignore file, again using a specific prompt and Junie performed this task without issues.
Unit tests followed that success, but something strange happened here with GoLand reporting 0% test coverage. Upon further investigation Michelangelo noted that the tests, while present, were missing the all important CoversClass attribute. Michelangelo had to iterate a few more times to get the tests to the state he requested - something he reflects happens with junior developers too. More prompt iteration followed and Michelangelo got his code coverage to 100% without any risky tests. I do wonder what the time saved here was, that would be an interesting experiment!
Michelangelo concluded his experimenting with Junie at this point. What about you? You can check out Junie too.



