Hey @pland !
I can’t comment on codeguide, as I’ve never used it before. I can definitely confirm on the benefits of documentation though.
The models like Claude and GPT have been trained on vast amounts of existing code, but every new project is a little different. What this means in practice, is that without guidance, their output will gradually drift away from what you wanted, towards what the model thinks you might have wanted.
This subtle and unhelpful rewriting of your project plan can really derail things, so it helps to be continually reminding the model what you’re trying to achieve. This is where docs can really help. If you have a spec document you can tag into the context for your prompt, you’re effectively enriching the prompt with lots more helpful information about what’s already determined, and where you’re going with it.
It can also help on larger projects to avoid things like “ah you need one of these functions” and the model writing it for you, when a perfectly good implementation already exists – it just didn’t notice.
I’ve worked on a lot of projects of various sizes and I always find that once the codebase starts to grow, having at least some documentation is pretty much essential.
If you’re just getting started, or starting to build larger projects, check out some of the tips in An Idiot’s Guide To Bigger Projects – it’s a few months old now (new version coming soon!) but you may find some of the recommendations there helpful.