Junie did mostly a great job, but where it failed miserabily was, when I asked it to integrate django-templated-email-md. I did a subtle typo and just asked for django-templated-mail-md. The result is, that it created disfunctional code that almost looked like it could work and followed the README of the correct project. But the main import included my typo and to solve the tests, Junie mocked almost everything to run tests successully.

Just checked, Claude Code did the same.

Done for today. My AI quota is almost used for this month. I did the same tasks over the evening with claude code too. Effectively this experiment costed $15, so it is fair that my monthly quota with Jetbrains is consumed. Jetbrains AI costs $10 on the Pro plan.

I have to review the docs tomorrow, then I am actually ready to release this library.

@t_redactyl I have an AI Pro subscription. How many tokens are included within this package? S, M, L are nice information for t-shirts, but for a software package it is quite abstract. I ask because PyCharm just told me, that I will have consumed my complete usage limit soon and I jused the feature today.

Junie had some kind of outtage in the last hours. Now we are back on track and apply the finishing touches. I think, I can complete this today. #django #pycharm

I excepted that the agent was able to setup a proper action for GitHub, but that it was aware of Codeberg and Forgejo surprised me. Very handy. Saved me the time to read the documentation. ;)

https://codeberg.org/oliverandrich/django-solomon/src/branch/main/.forgejo/workflows/tests.yml

I had mentioned that I prefer detailed and well-structured responses when requesting a proposal or suggestions from an agent. Junie deviates from the pattern to compress lengthy text into a sidebar. Instead, you receive a summary and a link to view the markdown document in a editor.
#django #pycharm

But for simple prompts prompts like the following, it could be a bit faster to be honest.

> Commit and push the changes.

Basically Junie always does the best job of the agents I tested so far. But the speed is quite slow. The reason for this is obvious, it always analyses the situation, runs the test suite and iterates through changes. It creates way less errors or broken tests compared to others.

Pretty happy with the rate limiting Junie implemented. But what I don't enjoy it, is that it always needs a reminder to achieve for 100% test coverage. Something that I would expect from a AI Agent. Especially, if it forgets signifikant test cases, it generates afterwards.

https://codeberg.org/oliverandrich/django-solomon/commit/a2fe5b541fe6214e92c17c0845a3b04c0d893910

Next iteration. I played stupid and ask for some proposals how to enhance the token security. I received 10 proposals including implementation details and benefits. As a good client, I asked for 2 out of ten. ;) Of course the proposals included what I would have done myself.

Junie has one weakness in Django projects. It tends to forget to create a migration on the first run. Sounds fixable. I create a ticket for it.

Of course Junie added changes that violated some rules of ruff. Even though these are no errors, but they provides better code quality. I asked Junie to commit and push the changes. This triggered the pre-commit hooks and Junie parsed the ruff output and fixed all issues. Nice.