Build a working tool with no code at all.
You have a hundred little tasks you do by hand or half-solve in a spreadsheet. This lesson turns one of them into a real, working tool using AI — no code, no setup. You'll learn what's worth building, how to describe it so AI gets it right, and finish with something you'll actually use.
The mental model
An "app" is just a tool that does one job. Describe the job clearly and AI builds it — you never touch code.
Forget "software development." A tool takes some input, does something useful, and gives you a result — a tip calculator, a quote estimator, an intake form. AI is remarkably good at building these from a plain description. Your job is to describe the job, not to engineer it.
Step 01 · What you can actually build
Good first tools share a shape: one clear input, one clear result, used over and over. Pick the kind you want to build — it sets the example in your prompt below.
Step 02 · Describe the tool in four parts
Before you prompt, nail down four things — this is the whole skill. Pick one option per row (use your own answers for real) and watch the build prompt assemble:
Step 03 · One job, or overbuilt?
The skill behind every good first tool is scoping. Tap each idea — is it a clean one-job tool, or already overbuilt for a first try?
Step 04 · When is a tool "done"?
- It does the job without you explaining how it works.
- Inputs are obvious; the result is impossible to miss.
- It doesn't break when someone does something unexpected.
- You'd actually use it instead of the old way — the only test that matters.
Your challenge: replace one manual task
Pick something you do in your head, on paper, or in a fiddly spreadsheet. Then:
- Write the four-part description (job, inputs, output, who uses it).
- Run the build prompt and answer the AI's questions.
- Do two refinement passes until it's clear and hard to break.
- Use it for real once — does it beat the old way?
Lesson complete
You can tell a one-job tool from an over-built app, describe a tool in four parts, generate it from a single prompt, refine it with plain feedback, and judge when it's done.
Lesson 2 · From toy to tool — data, logic & screens