Document Management

Problem

AI project can and should generate a lot of documentation: What's the architecture? How do you use it? How was a problem solved? What's left to do? What's the database schema? What are the APIs? ...

It's not uncommon, in a fairly large sized project, to hit around 100 different files. How do you make sense of them? If you ask it to summarize the other documents, now you just have one more document. How do you view the existing ones? Search through them?

You could, perhaps, move all the documents to something like Confluence. That would work, for an instant in time, but because you AI is going to want to keep updating the documents, the copy in Confluence will grow out of date, quickly.

Instead, this is my approach...

Structure

You'd do best to start with a "docs" directory in your project, somewhere, that will collect all the documentation that's generated. Then, divided that up into subdirectories like:

• user
• developer
• specifications
• tests

You need to instruct you AI coding agent to honor your structure when it creates new documents. If you start to collect a lot of documents, you can create futher sub-subdirectories to organize things.

Serving Documentation

Now that you have your documentation organized, how do you easily access it?

I use a program called mkdocs-material, which turns your docs directory into a searchable website with no effort. The searchable part makes it great. You can add structure if you like, but fundamentally it just looks through the directory and constructs a website out of it on the fly.

This Vibe Coding Cookbook is created this way.