How I work

My method has four moves.

Research a lot. Before I build anything, I study the problem from public sources: the platform's documentation and changelog, how other products solve it, and what real users say in reviews and forums. Every fact keeps its source.
Decide. I write the decision down, with the reason, so the next step doesn't depend on anyone's memory.
Anticipate the problem. I look for what is going to hurt the client before it happens: a deadline, a silent failure, a support queue that will fill up.
Solve it, and prove it. I build the fix and test it against the real system, not against a mockup. Then I document it so someone else can run it without me.

That is why I built the Embed Migration Kit before our call. The research showed three Shopify dates coming at once:

So I built the tool a team would need to move stores safely, and tested it on a development store.

How I run it day to day

I run a one-person studio that ships production software with AI agents under a written method. I direct, decide and verify; the agents execute.

One case file per project. Seven sheets: who decides, current state, what is pending, decisions, changes, the roadmap and the links. A new session opens the file and continues where the last one stopped.
Phases with approval gates. Nothing moves to the next phase without an OK, and I can give it from my phone.
Every change checked against the real system. Images are built without cache and compared with the tested code before they go live.
Guardrails. Agents cannot delete or overwrite anything that already existed, and they cannot publish client data, without my explicit OK. A silent mistake that happens twice becomes an automatic check.
A weekly log. Every change is checked against git and the database, with the real time it took.

These numbers were counted on my server on October 10–11, 2026.

58 Docker containers running (60 defined), behind one reverse proxy with automatic HTTPS.
34 public hostnames on 2 domains.
48 alert rules loaded in Prometheus.
118 Playwright scripts across two QA harnesses. Demos are checked in three browser engines before they go out.
22 scripts for my working method and 11 written playbooks for recurring jobs.
A complete PWA in 1 to 7 hours, the same day, often starting from a mockup or one of my templates.

I had not built on Shopify before this weekend.
Most of the hours behind this server are agent hours. Mine are in the hundreds, not thousands, and that ratio is the point.

The full version, with the evidence for each claim, is How I Ship: How I Ship

The method also fits in one drawing: see How I work, in one diagram (diagrama/diagrama-en.html).

Tap a stage to see what happens there and what it looks like in practice.