Healthy path · Step 3 of 4 (night one) · ~30–60 minutes

Your first session

Goal: finish something small you can click and explain — not a mystery mega-app.

Beginners often open an AI tool, type “build me an app,” burn credits (money), get confused, and quit.

This page is the antidote: one small thing to build → clearer prompt → one sandbox try.

Before this step

You should already know the idea ( Step 1 ) and have picked one tool ( Step 2 ). If not, do those first — then come back. Don’t jump to the lab until you’ve tried this once.

Quick reminder

Keep this page (or the printed checklist) open. Open your tool in a second tab. Do section 1 first — then signup / empty project / one prompt. Full two windows advice (always on Step 2, quiz or not): Two windows — don’t lose this guide.

1. Before you open the tool (5 minutes) — do this first

Stay on this page (or the printed checklist). Do not open Lovable / Cursor / etc. yet.

Block 30–60 minutes if you can.
Write down what you will build in this session — one short sentence. Not “an app that does everything.” Something small you can finish and click today. If you cannot say it in one sentence, it is still too big — cut half of it.
Example one-sentence builds optional · steal or adapt
- “A single page for my bakery: heading, short blurb, Contact button”
- “A todo list: add items, tick done, still there after refresh”
- “A tiny game: move, one obstacle, score goes up”
Write three must-nots — hard “no” lines for this try only. Why: left alone, the AI often adds stuff you never asked for because it is trying to look complete. Must-nots stop that bloat so you finish one small thing. Three is enough.
Example must-nots (pick any three) optional · stop AI bloat
- No user accounts / “sign up” screens
- No fake stats (“1,000 happy customers,” fake charts)
- No payments or “buy now” checkout
- No dashboards, admin panels, or “also add AI chat”
- No real email lists, passwords, or private notes pasted into the chat (even on a practice project)
Note free-tier limits — credits often mean money when they run out. Glance at free limits before you start; do not stop mid-build to study every paid plan page.
Plan a new empty project / folder ( sandbox ) — not important work.
Optional: copy the prompt from section 3 into Notes first, then paste once when the tool is open (less tab-flipping).

2. The hour (outline only — don’t start yet)

Preview only — not the do-now list. Finish section 1, read section 3, then open the tool and run this clock.

When you start the clock: tool in a second tab · this page (or checklist) still open · paste once · if stuck, re-read this guide — not ten other blogs.

0–10 min Open & paste once

New empty project. Paste your prompt (section 3, or a lab challenge).
How optional
- Second tab = tool; keep this guide open
- One paste only — don’t re-roll forever
- Stuck writing? Use a lab brief instead
10–25 min One full click-path

Stay on that one small thing until you can click it start → finish. No extras mid-way.
How / example optional
- Example (todo): add item → tick done → refresh → still there
- Do not ask “also add a dashboard / AI chat / more pages” yet
- Pretty but half-broken = keep fixing the path, not new features
25–45 min Fix one problem at a time

Click around, note what fails, tell the AI only that bug. Check a narrow window for phone layout.
How optional
- Write the failure in plain words (“add button does nothing,” “text cut off”)
- Short follow-up prompt: only that issue — not “make it better”
- Phone check: shrink the browser window narrow (or phone preview if you know it)
- If things overflow or overlap, ask for a simple phone-friendly layout
45–60 min Rough clicks, then stop

A few rough clicks (smoke-check), jot what works / what’s shaky, close the tool. Session over.
How optional
- Leave a field empty and hit submit
- Refresh — is your data still there?
- Click the same button twice
- Write a few short lines: works / still confusing
- Close the tool tab — no second mega-feature “while you’re here”

3. What to type (copy and adapt)

Vague prompts wander. Use three labels so nothing important is missing:

CONTEXT — who you are and what “done” means today
SCOPE — only what to build (the one page or list)
CONSTRAINTS — your must-nots (no accounts, no fake stats, …)

Paste the whole block into the tool (Lovable, Cursor, etc.). Edit the words to match what you’re building.

First session · todo list example

CONTEXT I am new to vibe coding. Today I only need a simple todo list I can use on my phone.

SCOPE One page: add items, tick them done, keep the list after refresh, show a friendly message when empty. Works on a phone-width screen.

CONSTRAINTS No sign-up. No fake stats. No payments. No dashboards. Keep it simple. At the end, list anything I should double-check.

Prefer a ready challenge instead? That’s Step 4 · Prompt lab — after this first try if you can.

4. Open the tool (new tab) and paste once

Only now — after your one-sentence goal, must-nots, and prompt are ready. Keep this page open. Second tab = tool. New empty project only. Paste once, then run the hour clock above.

Most people on this path use Lovable

Browser tool: sign up free if needed → new empty project → paste your prompt. You can use another tool you already chose — open that one instead.

Open Lovable (new tab) Lovable guide on this site

Open Lovable may be a tracked affiliate link — see disclosure. You do not pay extra.

5. If it goes wrong

Common snags. Each line is: what you’ll see → what to do.

The AI changed a pile of things at once and you can’t tell what’s going on

Stop accepting more big rewrites. Use Undo in the tool if you have it, or start a new empty project and paste a smaller prompt. Next message: ask for one fix only (e.g. “only fix the add button”).

It looks nice but buttons/forms don’t actually work

Tell the AI to drop the fancy extras and make the main action work (add / save / next). Ignore polish until the click-path works.

You pasted a password, private note, or secret code into the chat

Don’t do that again — even in practice. If it was a real password or API key, change it on the real service (new password / new key). Treat the chat as public-ish.

It looks so good you want real customers on it tonight

Not yet. Pretty is not the same as safe. Use it as a demo only until Step 5 · later.

Done for today when…

If most of these are true, Step 3 is a win — even if the result is only “partial.”

You can click through the main thing you built (even if it’s rough)
You know which tool you used (e.g. Lovable) and that it was a practice project
You jotted a few notes: what works, what’s still confusing (doesn’t need to be perfect)
You stopped — no “one more feature” spiral that burns credits (money) on polish

Next when you’re ready: Step 4 · Lab for another small practice, or rest. Production is later.