How to prompt an AI builder like a product manager

When an AI build goes sideways, the prompt is usually the culprit — not because it was too short, but because it described a screen instead of a problem. Product managers deal with this every day, and their fix is a good brief. The same habits make an AI builder far more useful.

Start with the outcome, not the screen#

“Add a form” tells the agent what to draw. “Let clients book a haircut without calling the salon” tells it what success looks like — and leaves room to propose the right flow: a service picker, a day and time selector and a confirmation step. Outcome first, layout second.

The five parts of a good brief#

1. Outcome. What should someone be able to do when this is finished?
2. Audience. Who is it for, and what device will they use?
3. Scope. The pages, data and actions involved — and what’s explicitly out of scope for now.
4. Constraints. Brand, tone, existing content and anything that must not change.
5. Definition of done. How you’ll check the result in the preview.

A template you can copy#

Goal: Let clients of Lumière Salon book an appointment online.
Audience: Existing clients, mostly on their phones.

Scope:
- Services: Haircut, Color, Styling
- Pick a day, then an available time slot
- Confirmation screen with a booking summary
Out of scope for now: payments, staff accounts.

Constraints: Match the site’s colors and fonts.
Keep copy short and friendly.

Done when: I can book a Color appointment on Thursday
at 10:30 in the preview and see the summary.

Lumière Salon is just an example — swap in your own business and keep the structure.

Scope small, iterate fast#

Big-bang prompts produce big-bang changes that are hard to review. Ask for one slice at a time: the booking flow first, then the confirmation step, then the view your team uses to manage bookings. Each change lands in the live preview and is saved as a checkpoint, so reviewing is quick and rolling back is painless.

Give feedback like a reviewer#

• Point, don’t describe. Annotate the preview or @mention the file you mean instead of explaining where something is.
• Name the gap. “The time slots should wrap to two rows on mobile” beats “fix mobile”.
• One change per message when it matters. It keeps checkpoints meaningful and easy to roll back.
• Say what to keep. “Keep the layout, change only the copy” prevents unwanted rework.

A good prompt is a good brief: it says what done looks like and leaves the how to the builder.

Use the code view as a spec check#

You don’t need to write code to benefit from reading it. Flipping to the code view shows how the agent structured a feature — which components exist and what data they hold — so you can catch a misunderstanding before it spreads. And because it’s real, framework-standard code, a developer on your team can pick it up at any time.

Try the template on your next idea in a free workspace, or read more about how the AI builder works.