Learning to be the engineer again

The desync between the different systems could easily be chalked up to AI issues, as the codebase is largely written by Codex; however, that is not the sole reason.

Throughout the development of this project, the codebase has been through two major iterations. The prior iteration—Snapshots 0.1.0 to 0.6.0—is just an abhorrent disaster. Barely anything worked properly, but it proved that a project like this was possible.

For 0.7.0, I started anew, with plans to try to write the majority of it by hand. The core functionality was strong, the initial version of the new project wizard worked wonders, and I even managed, with some assistance, to get Blender to launch. However, quality started to slip when I stopped being the engineer and just started delegating everything to Codex.

I’ve started to take a more hands-on approach with Codex and even try to write the code myself. However, it’s clear that my inexperience with Kotlin and the IntelliJ API is the real bottleneck within the codebase. Granted, IntelliJ’s docs are absolutely atrocious and often just throw you straight into a file somewhere in the IntelliJ Community repository, which more often than not is written in Java on the older IntelliJ Platform Gradle Plugin v1.

Because of this, I’ve stopped relying on the IntelliJ docs and started using Codex a lot more. For the most part, it is very helpful—especially for automating tasks that take forever, including documentation updates and minor yet tedious refactors—but it can still produce some issues.

And, of course, when I was using it via my JetBrains AI Credits, it was expensive. I recently switched to using a ChatGPT subscription, which has already proven beneficial, as I have easily gotten more usage of Codex via that subscription than by buying top-off tokens from JetBrains. Unfortunately, those are only really worthwhile when using Junie, since Junie doesn’t just gobble them up like me when I have a bag of gummy bears. But the quality and output, like those gummy bears, are often shoddy at best.

However, that being said, the automation—mainly in regard to publishing, Git commits, and changelogs—has been incredible.

Now that’s out of the way: here’s to good things. The future of Blender Development for PyCharm is looking wonderful!