Course A · Module A4 Build · Lesson 13 of 17
The one page you write before touching any AI builder is the page that decides whether your build actually works.
Most first builds fail before the first prompt. Not because the AI tool is bad — because the founder never decided exactly what they were making. They paste a vague idea, the AI guesses, the result is 70% right but 30% wrong in ways that are hard to fix, and two hours later they're rewriting everything. Your single win today: write a build brief — a one-page plain-English spec — before you open the builder. It is the biggest lever on output quality, and it takes 15 minutes.
A contractor who builds exactly what their client describes is only as good as the description. Hand them "a nice kitchen" and you'll get a generic one. Hand them "a galley kitchen for a caterer: 12 ft run, double sink left of the range, 6 burners, no island" and you get something useful on day one.
Your build brief is the contractor's blueprint. The AI reads it once and makes decisions for every line of code it writes. Vague brief → vague app → expensive rework. Sharp brief → focused app → faster iteration.
The "leave out" column isn't permanent — it's version 2. The number-one mistake in a first build is scope creep before you have anything working. One narrow job, done well, is a win. A half-finished multi-feature app is not.
Copy this into a doc or paste it directly into your AI builder as the opening of your prompt. Fill every blank before you proceed. If a blank feels hard, that difficulty is useful — it means you haven't decided yet, and it's better to decide now than at 11 pm staring at broken code.
Here is a completed brief so you can see the bar to clear:
Notice: no login, no history, no email — just the one transform. That's exactly what you paste into Lovable or Bolt next lesson. This brief is your A14 starting prompt.
Answer before opening — recalling it is what makes it stick.
It contains two jobs. "Generates invoices" and "emails them" are separate features. Pick one for v1 — most likely the generator, since that's the harder UI piece. Add email sending in v2 once the invoice form works.
It tells the AI what NOT to build. Without it, the builder may helpfully add login, history, or settings — all of which add complexity and breakage before you even have the core working. Explicit exclusions are as important as inclusions.
It's untestable. "Done" must be a concrete, runnable test: a specific input that produces a specific output you can check right now. Vague done-criteria mean you'll never know if you're finished — and neither will the AI when it evaluates its own output.
Read this next (primary source)
Prompting best practices — Lovable Documentation. The official guide to writing prompts that produce focused, working apps — the same principles as the build brief, applied inside the builder. ~10 min. See RESOURCES.md for more.
New words this lesson — build brief, scope — are defined in the course glossary. Keep it open.