Tinkerer

You've written scripts, wrangled a website, maybe automated something at work, and AI has made you dramatically faster. This guide is about turning that speed into software that holds up: versioned, configured properly, and reviewed before it ships.

Not quite you? The Explorer guide starts from zero; the Developer guide is the rules-only digest.

The core workflow

The single biggest upgrade from “prompt and pray” is a rhythm. Ours has four beats, and it scales from a weekend project to production software:

1. Spec

   Write what you're building in plain English before any code — a paragraph or two. Paste it into the chat first, and keep a copy in your repo. Then ask the AI: “What's missing? What would a senior developer ask?”

2. Architecture

   Ask for two or three ways to structure it, with trade-offs. Push back. Ask what breaks if usage grows tenfold. Ten minutes here saves hours of rework — and you make the call.

3. Phased prompts

   Build one coherent slice at a time — data, then logic, then interface, then tests. Each slice is a natural point to commit your work and review before moving on. Don't let an agent run unsupervised for hours.

4. Review

   Read what the AI wrote before you keep it. If a change is too big to read, it's too big to accept — ask for it in smaller pieces. Plausible is not the same as correct.

The full workflow, with examples

Two habits to build this month

Reading what the scanners tell you

Once your projects live in GitHub, automated tooling starts talking to you: secret scanning, dependency alerts, security findings. Don't ignore them, and don't panic either. Paste a finding into a fresh AI chat and ask what it means, whether it applies to your code, and what the fix looks like. Learning to read findings is what makes an app trustworthy enough for real users.

Where to go next