Motive

The Philosophy Underneath

Let it develop naturally

i like that we let structure emerge, like a journey without a map. Start with nice pithy articulation of raw curiosity. Travel carefully. The structure comes at the end.

Work files are just that — things we need to know and things we did. Guides are the maps of the turf, what's been learned about it. The guides are condensed out of the work, things like "oh, shit, we need to do this and not that."

Dual purpose

Documentation has two audiences: me, a lot later and rather forgetful, and AI the amnesiac.

Work tracking is the other side of the coin. Focus, details, decisions. Later, read the file, know exactly where we are. My brain feels better just saying this.

Small, composable pieces

Files stay focused. It might be a class, it might be a concern. When something grows too big, split it. And, jeez, the fewer the better. Shrink, evaporate, snip. Some lessons are universal — they get promoted to shared.

Friction into feature

We describe our process, too. Like, "ugh, this thing is badly broken." Pause and scratch noggin. Often enough something cool settles into play.

Patterns come from pain — why else remember them?

How It Started

My first uses for AI were one-off: gee, how do you do such-and-such? I'd take the advice, often asking for more. Over time, I noticed I was asking the same things repeatedly because I wouldn't remember the answers. So I started asking AI to write concise descriptions of what we did together, in markdown.

The pile grew unmanageable. I asked AI to help: find redundancies, tighten the writing, make it easy to use. What I have now: clean separation of purpose.

Then I asked if these markdown files could be published as a static website. BOOM — up-to-date documentation, something I have NEVER before encountered. And more: a shared context that lets AI pick up exactly where we left off.

Why This Works

Guides encode decisions once — no re-explaining, no drift. Work tracking survives sessions. Documentation is dual-purpose: helps future-me and brings co up to speed instantly. Most AI interactions are one-off Q&A. This is collaborative project work with accumulated context. A different game.

What i still own: taste, creative direction, judgment calls. Co doesn't feel when something's "too gappy" — co follows rules once i articulate them.

Is This Approach Common?

Not really. Custom instructions, .cursorrules, Claude Projects — similar pieces exist, but usually static personality tweaks or code style, not project-specific workflow with accumulated wisdom. What's less common: guides that evolve from actual pain, work tracking with resume points, treating documentation as shared context rather than just human reference.

This feels more like pair programming methodology than prompt engineering. An institutional memory system that happens to include an AI.