I put up a new AGQ pair today around a workflow habit that has quietly become one of the most useful parts of how I work with AI on larger projects.
Guide:
https://askgoodquestions.dev/guides/keep-the-build-thread-clean
Field note:
This pair is about separating active implementation from feature exploration instead of trying to make one AI thread carry both.
The basic pattern is simple:
Keep one thread focused on the current build.
When a substantial side issue comes up, move it into its own focused conversation.
Then turn the useful result into a markdown handoff and promote accepted decisions into the repo so they become part of the project’s actual memory.
What I’ve found is that this does more than keep the main thread cleaner. It also lets you work ahead. You can think through future features, sort out tricky decisions, and have those documents parked and ready until the active build reaches the point where they belong.
That has been especially valuable because AI can look like it has absorbed a giant prompt or giant design document when in reality it has only skimmed enough to feel oriented. Breaking things into focused decision files makes it much easier to bring in the right context at the right time.
The guide lays out the workflow and the reasoning behind it.
The field note is the story side of it and how this has started to become a real operating method for me rather than just an occasional trick.
Main takeaway:
The better you structure the work, the better the AI tends to perform.
I’m curious whether others here are doing something similar, where the project repo becomes not just the source of truth for code, but also the source of truth for accepted design decisions.
Charles