I don’t wing this. Here’s the actual system.

Before I build anything, I get interrogated.

Every project starts with a skill I run called /grill-me — a relentless interview that pokes holes in the idea before a single line of code exists. It asks the questions I’d rather skip: who this is actually for, what happens when it’s wrong, what I’m assuming that isn’t true yet. Half my best decisions happen here, before anything is expensive to change.

I don’t guess at a design once. I generate five and argue with myself.

/design and /ui-ux-pro-max turn a brief into real, comparable options instead of a single first draft — different layouts, different type systems, different color logic, side by side. Most of them get thrown away. I’ve never once regretted seeing the version I didn’t build.

Before anything ships, I check whether Google will actually find it.

/seo-audit runs the technical and content checks I’d otherwise forget under deadline pressure — indexability, structured data, thin-content risk, the stuff that quietly tanks a launch. It’s not theoretical: one product I run this way has pulled over a million pageviews since launch, and that’s not an accident.

The actual edge, though, was never the AI. I curated real, specific data most people can’t cleanly get their hands on, then used AI to process and structure it faster than I ever could by hand. The AI didn’t invent the advantage — it scaled one I already had. That’s the part worth copying, not the tool name.

Every project keeps a memory. So does the AI working on it.

I run a memory.md file per project that Claude reads and updates as it works — current state, decisions made, what’s still open. Come back after a week away and it can tell you exactly where things stand in one paragraph, instead of you re-reading a week of commits to remember your own project.

Claude doesn’t get to freelance.

Every project starts with the same non-negotiables: never push to git without asking first, always ask when something is genuinely unclear instead of guessing, and back-test real changes with Playwright or end-to-end tests before calling anything done. The AI is fast. These rules are what keep fast from turning into reckless.