A launch waitlist for a pickle brand
The smallest useful thing: a page saying what's coming, with a form that stores signups somewhere you can actually read them.
You're launching a food brand on Instagram and want a real link for the bio this afternoon.
Start with the whole thing, not the first piece
The instinct is to ask for a page first and the form second. Don't. The agent plans better when it knows where it's going, and you get something you can look at sooner.
Build a launch page for Aai's Achar, a pickle and masala brand from Nagpur — my mother's recipes, made in small batches. It needs a headline, a short description, and a form that takes a WhatsApp number, with an email box for anyone who prefers that. Promise everyone on the list a discount in launch week. I want to be able to see the list of signups.
Two things earn their place there. The WhatsApp field leads because that is where these customers are; email is the fallback, not the point. And the last sentence matters — "takes a number" alone gets you a table you have no way to read. Saying what you need to do with the data is usually the difference between one request and three.
What it did
It built the page, added a form posting to an API route, created a table for the signups — number, email, when they joined — and added a second page listing them. The checks passed on the first attempt and it went live.
The copy it wrote for the page was generic food-brand filler — "authentic flavours, made with love" — words that could sit above any jar in any shop. That is worth knowing up front: it will produce plausible filler text unless you give it your own words.
Then fix the copy, in your voice
Replace the headline and body copy with this, word for word: "Aai has made this lemon pickle every summer for thirty years. This year she made extra." Don't rewrite it or make it more polished.
"Don't make it more polished" earns its place in that prompt. Without it the agent will improve your writing back toward the mean, which is exactly what you were trying to escape.
What we had to correct
Every guide on this site has this section. If it were empty, the guide would be a demo.
The signups list was public. Nothing in the prompt said it shouldn't be, and the agent didn't ask. Anyone with the address could read every WhatsApp number and email address collected. This is the kind of mistake that passes every check we run — it is working code doing exactly what it was told, and it was told the wrong thing.
Fixing it took one request: "the signups page needs to require a password before it shows anything." But the lesson generalises. Our checks catch broken code. They do not catch code that works and is a bad idea.
It used the production database from the start. Every project gets a development and a production environment, and the agent defaulted to building against your working copy — correct — but the live page was reading from the same one until we said so. Worth checking rather than assuming.
Where it stopped short
No message goes out to anyone who joins — the launch-week discount is a promise on the page, and keeping it means sending those WhatsApp messages yourself, from your own phone. Nothing tells you when a new signup arrives — you have to go and look. And there is no export, so getting the list into anything else means copying it by hand or asking for an export in another request.
None of that is a limitation of the tool. It's a limitation of the four sentences we gave it, which is the honest general lesson: what you get is bounded by what you thought to ask for.