- Your first Project and file uploads
- The major AI families to try the workflow
- A small daily message allowance
Create an MVP with AI
Create an MVP with AI: cut it until it fits in three weeks, then build only that.
An MVP is not a small version of the whole product. It is the smallest thing that lets a real customer get the result they came for, so you can learn whether they want it. Most MVPs fail by being too big, not too small. This page is the cutting method, the prioritisation that survives contact with reality, and the first build plan, with several AIs doing the arguing so you do not have to argue with yourself.
Free to start. Scope decisions are judgement calls; the latest models make better ones.
The cut
Four passes that take a product idea down to an MVP
Name the result, not the product
"A small agency owner gets a polite reminder sent to a late-paying client without writing it." Not "an invoicing platform". If you cannot say the result in one sentence, you do not have an MVP yet; you have a category.
List everything, then sort into three piles
Everything you imagined the product doing. Then: needed for the result, nice after the result, and someday. Be honest: sign-in is often "nice after"; a logo is always "someday".
Ask three models to cut it further
Give each the result sentence and the "needed" pile. Ask each to remove half. Where they agree on what goes, it goes. Where one keeps something the others cut, that is a real decision; make it with the reasons visible.
Set the three-week test
If the remaining list cannot be built in three weeks of evenings with AI, it is still not an MVP. Cut again. The point is to learn, and learning is faster with less.
Prioritising what stays
A scoring that survives your own enthusiasm
| Question | Score | Why it matters |
|---|---|---|
| Does the result happen without this? | 0 if no, 3 if yes | The only question that defines an MVP. |
| How many of the ten interviewees asked for it? | 0 to 3 | Features one person asked for are wishes, not requirements. |
| How long will it take with AI? | 3 if a day, 0 if a week | Time is the budget. Expensive features need stronger evidence. |
| What breaks if it is wrong? | 0 if money or data, 3 if nothing | Anything touching money or data costs extra review time. |
The tie-breaker
When two features score the same, keep the one that produces something the customer can show someone else. Sharing is how MVPs find their second customer.
The build plan
Three screens, in the order the customer meets them
Almost every MVP is three screens: the one where the customer gives you something, the one where the result appears, and the one that gets them back. For the invoice example: paste an invoice, approve the reminder, see the status. Write the user flow for each as a numbered list, in the Project. Then build them in that order with the screen-by-screen loop from the app guide, with a second model reviewing each before you move on.
Data can start on-device or in a spreadsheet. Sign-in can be a magic link or nothing. Payment can be an invoice you send by hand. Every one of these is a real product decision you can defer until a customer makes you decide it.
Free is the start. The latest models are the difference.
Cutting scope is the highest-leverage decision in the product. Make it with the best models.
Older models agree with whatever feature list you give them. The latest models, asked to remove half, argue for what to remove and why, and they disagree with each other usefully. An MVP that is one week smaller because of a better argument is a week sooner to learning whether anyone wants it.
Newer models hold your whole Project in view instead of the last few messages, reason through trade-offs instead of picking the first plausible answer, and are wrong less often and with less confidence. For the questions on this page, that is the difference between advice that sounds right and advice you can act on.
Pro from $20 a month, cancel any time. The free plan stays free.
- The latest ChatGPT, Claude, Gemini, Grok and Perplexity models, all on the same Project
- Far more messages a day, so a working session does not stop halfway
- Modes where several models draft, challenge and verify each other's answers
- More Projects, so each MVP experiment has its own context
- Documents and decks generated from the Project for design partners
Start here
First messages for an MVP Project
Add the product idea, the interview notes if you have them, and the full wish list. Then start here.
ResultState the single result a customer gets from this product in one sentence. If you cannot, tell me why and what I need to decide first.
PilesSort this feature list into three piles: needed for the result, nice after, someday. Explain any that were hard to place.
CutRemove half of the "needed" pile. Argue for each removal. Then tell me what you would keep even if I pushed back.
Three weeksCan this be built in three weeks of evenings with AI? If not, what else goes?
FlowsWrite the user flow for the three core screens as numbered steps, from the customer's point of view.
Build planTurn the flows into a build plan: which screen first, what data it needs, what can be faked for now, and what the customer will see at the end of each week.
Honest answers
What people ask
How small should an MVP be?
Small enough to build in about three weeks of evenings with AI, and large enough that one customer gets the result they came for. If those two conflict, the idea needs a narrower customer.
Should the MVP have sign-in and payments?
Usually not at first. A magic link or nothing for sign-in; a hand-sent invoice for payment. Add the real thing when a customer makes you.
What if customers ask for features during the MVP?
Write every request down with a tally. Build only what several unrelated people ask for. Single requests are wishes.
How do I know the MVP worked?
Someone uses it without being reminded and one of them refers someone else. Not sign-ups; use.
Why several AIs for scope decisions?
Because one model agrees with your list. Three models asked to cut it disagree in useful ways, and the argument is what you need to hear before you spend three weeks.
What is free?
Your first Project, file uploads and the major model families. The latest models and more Projects are on paid plans. See plans.
Write the result sentence. Then cut until it fits.
Put the wish list in a Project and ask three models to remove half. Free to start.
Also: build a SaaS with AI · build software with AI · validate the idea · launch it