I don't just prompt AI. I design the system it works inside.
A prompt gets one good answer. Real work needs the same good answer every time, on every project, for anyone. So I design the environment around the AI: written rules, project memory, checks and hand-offs.
- Role
- System designer
- Type
- AI working environment
- Parts
- Skills, memory, checks
- Used for
- Every product on this site
Prompting doesn't scale.
Every new chat starts from zero. You explain the project again, the same mistakes come back, and what worked once can't be handed to anyone else.
Your colours, file names and last week's decisions don't carry over.
Ask twice, get two answers. A new screen won't match the last twenty.
What worked once, by luck, can't be handed to another project or person.
From writing prompts to designing a system.
The question changed from “what should I ask?” to “how should this work go, every time?”
Five parts, each one written down.
The skill says how the job is done. The memory says what this project already decided. Nothing ships until the check passes.
Three real projects. Each has its own memory, in the same shape: the rules, the decisions already made, and how to check the work.
- 00_START_HERE
- 01_CONVENTIONS
- 02_MEMORY_SYSTEM
- 03_PROJECTS_INDEX
- PROJECT_TEMPLATE
- 01_Memory
- 02_Skills
- 03_Agents
- 05_Assets
- 06_Tools
- 01_working_rules
- 03_design_system
- 04_copy_style
- 05_decision_log
- 06_verification
The project changes. The system stays.
Three very different projects run on the same skills and the same checks. Only the memory is different, so adding a project costs almost nothing.
A skill that turns an HTML screen into a real Figma frame.
Editable, with Auto Layout, not a picture. Every mistake it ever made is now a written rule inside it.
New frames come filled with white. Clear every one.
Set the width first, then let it wrap.
The tool doesn't have one, so it's written by hand.
Each of these mistakes cost a day once. Now they're rules, so they can't cost another.
This portfolio was built the same way.
Its working rules, design system, copy style, decision log and checks are all written down.
Any new session reads them first and continues exactly where the last one stopped, with the same standards.
A system is only worth what it ships.
Six products went through it. Each one kept its own memory and passed the same checks.
Don't make AI remember everything. Build something that tells it what it needs to know.
What I'd carry into the next one.
AI is one part
It builds. It doesn't decide what's worth building, or whether the result is good.
Narrow beats general
One job per skill. A general skill guesses, and guessing caused the mistakes.
It needs upkeep
A rule goes stale when the product changes. I update it when something breaks.
