I’ve been seeing a lot of discussion about AI tools like Claude Opus and ChatGPT changing how software is developed. I’m curious how this applies to a complex project like this one.
Do contributors here actually use LLM-based tools in their workflow? If so, in what kinds of tasks do they help (e.g., debugging, boilerplate, documentation), and where do they fall short?
I’m especially interested in whether these tools make a meaningful difference in such a large, long-lived codebase, or if their impact is mostly limited to smaller or more standard tasks.
If LLMs disappeared today, it wouldn’t affect PixiEditor’s codebase almost at all.
Sure, I sometimes ask ChatGPT about some algorithmic stuff that I am not really familiar with, its useful in that area, but if not the LLM, I’d just google it.
It does help a little with writing boilerplate stuff too, but honestly bigger impact on the editing speed was learning vim in my case.
It’s really crucial that I actually understand what’s happening in the codebase as PixiEditor is a really complex and advanced project. Letting any agentic tool touch the core codebase would be asking for problems.