Plan the smallest real product before you prompt.
An MVP is not the thinnest demo. It is the smallest version that lets a real user complete the real job.
AI builders make it easy to add features before you have decided the product. That is how a simple idea becomes a crowded dashboard, a half-working app, and a weekend of re-prompting.
Before you prompt, plan the smallest real product. Not the smallest screen. Not the cheapest mockup. The smallest thing a real person can use to get the core job done.
Before prompting an AI builder, plan the smallest real product: one primary user, one core job, one complete journey, clear feature cuts and acceptance criteria. That gives the tool a product spine instead of a feature pile.
- Pick one user and one job for version one.
- Sort features into Build now, Later and Never.
- Include recovery states so the first build is usable, not just impressive.
Start with one user
Most early products fail at scope because they are trying to serve three audiences at once. The founder wants customers, admins, partners, staff and future investors all represented in version one.
Pick one primary user for the first build. Write them as a situation, not a demographic.
"Busy professionals who want wellness classes."
"A returning client on mobile who wants to book their usual Tuesday class before it fills up."
The second version is not more detailed because it has more words. It is better because it tells the product what to prioritise.
Define the one job
The smallest real product needs one main job. If the user can complete that job, the product has a reason to exist. If they cannot, no secondary feature matters.
- Book a class.
- Request a quote.
- Create a project brief.
- Upload documents for review.
- Find the right expert and send an enquiry.
Notice the verbs. A product is not a collection of pages. It is a path to an outcome.
Sort the feature pile
Write every feature you want. Then sort them into three columns before you build.
- Build now. Required for the primary user to complete the primary job.
- Later. Useful, but not required for the first successful journey.
- Never for this version. Interesting, but likely to distract, slow or confuse the first release.
If removing a feature does not stop the main user completing the main job, it is probably not version-one scope.
Plan the full journey, not just the happy path
A thin MVP still needs recovery. "Small" does not mean careless. The first version should include the moments that make the core journey trustworthy.
- Empty state: what does the user see before they have data?
- Error state: what happens when an action fails?
- Confirmation: how do they know the task is complete?
- Change path: can they correct a wrong choice?
- Exit path: what if they are not ready to finish?
These are not "advanced" features. They are the difference between a demo and a usable product.
Write acceptance criteria
Acceptance criteria are plain tests for whether the build is right. They keep you from judging the product only by how nice the first screen looks.
Use AI after the shape is clear
Once you have the user, job, scope and acceptance criteria, AI becomes much more useful. It can produce screens from a real product shape instead of guessing which product you meant.
The prompt becomes less magical and more precise: build this user journey, with these rules, these states and these exclusions. That is where the tool starts feeling fast instead of random.
A small product can still be serious
Small does not mean basic. Small means focused. A clear booking flow, a clear quote request, a clear onboarding journey or a clear internal tool can create more trust than a sprawling feature set that never quite works.
That is the mindset behind a Product Direction Sprint or Blueprint Sprint: decide the spine first. Then build around it.
What should you plan before prompting an AI builder?
Plan the primary user, core job, main journey, feature cuts, recovery states and acceptance criteria.
What is the smallest real product?
It is the smallest version that lets a real user complete the real job, with enough recovery and confirmation to be trustworthy.
How do you cut MVP features before using AI?
Sort every idea into Build now, Later and Never. Version one only keeps what the primary user needs to complete the primary job.