How I build with AI
AI made building abundant. It didn't make judgment abundant. I use AI as leverage inside a system I own, not a replacement for owning it. Here's how that works, so you know what you're getting when my name is on something.
I'm not a career engineer. I'm a systems thinker who builds real things: a game, software products, this body of writing. Along the way I built a working method for doing that with AI without losing the plot. The method is the point.
What I use AI for
- Thinking before building. Pressure-testing an idea, arguing the other side, and surfacing the failure modes before I write a line. Not handing me an answer.
- Building within constraints I set. It generates the implementation. I own the shape, the boundaries, and the rules it has to respect.
- Research and synthesis. Gathering, summarizing, formatting. The groundwork, not the conclusions.
- Drafting from real work. Turning real notes and real projects into first drafts. The raw material is always mine.
- Auditing my own work. Running AI reviews against my own code and decisions to find what I missed.
What I don't hand off
Some things aren't the machine's to decide, and I don't pretend otherwise.
- Direction. What to build, what to cut, what it's for and who it's for. Mine.
- Risk. What to constrain, what to defer, what's acceptable to get wrong. Mine.
- Truth. Every story, number, and lesson I publish comes from real work. I don't let AI invent experience I didn't have.
The AI owns execution and a lot of the technical craft. It does not own the responsibility. Those are different jobs, and I keep them separate.
The part most people skip: the system
The reason this holds together isn't clever prompting. It's that the judgment lives outside the chat, in writing, where it can't quietly drift:
- A source of truth the AI reads before it does anything. The permanent facts, the decisions already made, and an explicit list of things it is not allowed to build.
- A log of every real decision, so a choice made once doesn't get silently reversed three sessions later.
- A rule that every hard-won lesson gets written down the moment it's learned, not left in a conversation that disappears.
Most AI work forgets everything between sessions. Mine remembers on purpose. That's what lets one person keep a large, AI-built system coherent over months.
Nothing ships on confidence alone
AI is fluent, which is not the same as correct. So nothing goes out on the strength of how sure it sounds.
- It runs in a staging step I test before anything reaches production.
- It passes a checklist: security, inputs validated, no secrets where they don't belong.
- I verify the actual experience myself. A model can't tell me whether something is good. I check.
The voice, and the ownership
AI helps me draft. It doesn't get to be me. I strip the tells: the hype, the filler, the tidy little symmetries that give it away. What's left should sound like someone who actually did the work.
On the field notes I make that split visible instead of hiding it. Each note carries a line on who did what: I direct the work and lived the decisions, Claude does the writing, I edit and approve every line. And where an idea or a reference came from Claude rather than from my own experience, the note says so on the page. Transparency about the method is the method.
If my name is on it, I own it. The product, the writing, the code. Its accuracy, its judgment, its consequences. AI-assisted is not AI-absolved.