Not a developer. Shipping anyway.
I've built a game, two apps, and a marketing system solo by running AI like a studio. These notes are what holds up when the tools change. Every one is extracted from a real project log, not theory.
New here? Start with note zero: what I am, what I'm not, and why the gap is the point.
The ledger
- Oct 01, 2026
The scheduled disagreement
The AI that helped you build it will not find what is wrong with it. One cold review with no memory of the work returned 47 findings, including two systems that had been quietly doing nothing.
Source: RL7 periodic-review.md, run 2026-06-21 - Sep 24, 2026
Nobody reads fifty files
Writing the lesson down is the cheap half. If nothing points at the lesson at the moment you need it, you will break the same thing twice with the fix already written.
Source: RL7 knowledgebase-index.md + context-management.md, 2026 - Sep 17, 2026
The mismatch is on purpose
An AI will sand the odd choices out of your work, because from inside a document a deliberate inconsistency looks exactly like a mistake.
Source: RL7 canon.md + drift-checks.md, August 2026 - Sep 03, 2026
Preflight checks for tasks, not just planes
A short fixed checklist I run before starting work, so I catch wrong-problem work and repeat mistakes before I write a line.
Source: RL7 task-preflights.md, in use since June 2026 - Aug 27, 2026
Predicting your own failure modes
A written list of how the project will probably break next, so I catch the predicted bug fast instead of rediscovering it slow.
Source: RL7 predictive-risks.md, maintained since early 2026 - Aug 20, 2026
The banned words list
AI writing has fingerprints. I keep a list of them and run it against every draft, including this one.
Source: Margin Harbor content scrubber + this site's voice guide, in use since 2026 - Aug 13, 2026
Two AIs, one project, zero collisions
Running two different AI assistants on the same codebase without them clobbering each other's work.
Source: RL7 dual-agent setup (Claude + Codex), in use since June 2026 - Aug 06, 2026
I can't read the code, so I test the behavior
What I verify instead of reading, what that catches, and, honestly, what it can't.
Source: RL7 playtest checklist + task preflights, in use since June 2026 - Jul 30, 2026
The file that lists what not to build
AI is relentlessly helpful. Without a written fence, that help walks your project somewhere you already decided not to go.
Source: RL7 drift-checks.md, in use since May 2026 - Jul 18, 2026
Your AI forgets everything. Build it a memory.
The file system that lets me run months-long projects with a collaborator that wakes up blank every session.
Source: RL7 project-memory system, in production since May 2026 - Jul 17, 2026
Verbs before varnish
The production rule that stopped me from polishing work I was about to throw away.
Source: RL7 production model, locked July 2026 - Jul 16, 2026
I don't know what I don't know
What I am, what I'm not, and the system that exists because of the difference.
Source: Site charter, July 2026