A startup MVP app can be built in a weekend with an AI agent if you scope it to one core action: describe the app in plain English on Saturday morning, iterate in the live preview, add a hosted backend with accounts so you can tell users apart, install the test APK on ten people's phones on Sunday, and measure whether they do the core action more than once. Export the source and hire a developer only when real usage says the idea works. The free plan covers a small MVP; the $1.49 Member plan adds unlimited builds and export.
The purpose of an MVP is not to have an app. It is to find out, as cheaply as possible, whether anyone wants what you are building. That reframes the weekend: the goal is ten strangers doing your core action by Sunday night, and the app is just the means. Below is a plan for the three days, the prompt that starts it, how to handle backend and testers, what to measure, and the signals that tell you it is time to spend real money.
Friday evening: scope the MVP to one action
Write down the single thing a user does that proves the idea. For a dog-walking marketplace it is "a dog owner books a walk"; for a study app it is "a student completes a session"; for a B2B tool it is "a manager approves a request". Everything in the MVP exists to make that action possible and observable.
- Keep: sign-in, the screen where the core action happens, the minimum data to make it real, and a way to see what happened (a simple owner dashboard or a list).
- Cut: payments (take a deposit by bank transfer or a payment link for now), notifications, chat, an iPhone version, a marketing site, an admin panel with filters, onboarding tours, dark mode.
- Fake: anything a human can do behind the scenes for ten users. Matching, approvals, pricing and support can all be you with a spreadsheet this weekend.
Write the scope as five bullet points: users, the core action, the three screens, the data, the look. That is your prompt. Founders who skip this step spend Saturday changing their mind in the chat instead of building.
Saturday morning: the prompt and the first hour
Open the AI app builder, create a mobile app project and paste the scope as a single message. Here is one for a local marketplace MVP; swap in your own nouns.
A preview appears within minutes. Spend the first hour tapping through it as an owner, then as a walker, fixing what is wrong in plain sentences: "the walker list should sort by price", "add a Cancel button on requested walks", "the founder dashboard should show a count of bookings per day". The login and dashboard guide covers the sign-in, roles and dashboard pattern if you want to understand what the agent built.
Saturday afternoon: hosted backend, second account, real phone
- Hosted backend. The prompt asked for it, which means accounts, a database and files run on the platform's servers and every tester sees the same bookings. If you forgot, one message adds it: "Move the data to a hosted backend with accounts so users share it."
- Roles enforced on the server. Ask explicitly: "Make sure walkers cannot see other walkers' requests and only my account can open the founder dashboard, enforced on the backend." An MVP with a leaky dashboard is an embarrassing Sunday.
- Two accounts, one flow. Sign in as an owner in the preview and as a walker on your phone's test APK, and run one booking end to end. This is where half the fixes come from.
- Empty states and errors. "No walkers yet", "Booking failed, try again". Testers judge an MVP on these more than on the logo.
- Icon and name. Five minutes, and testers take the app more seriously.
Use the preview's Simulate panel to test a bad connection and a slow phone before your testers find those states for you. A spinner that never ends is the number one reason a tester quietly gives up.
Sunday: the test APK on ten phones
Download the test APK from the preview header and send it to your testers as a file or link; it installs on any Android phone after they allow installs from their browser or file manager. Send the install troubleshooting guide along with it so you are not doing support all day. For testers on iPhone, ask the agent for a website version of the same app on the same backend; it opens in Safari and counts just as much.
- Recruit ten people who have the problem, not ten friends who like you. Two dog owners you met at the park beat eight cousins.
- Give them one task: "Book a walk for Tuesday." Do not explain the app; watch whether they can do it.
- Be the other side yourself. Accept every walk from a walker account so the loop closes today.
- Ask one question afterwards: "Would you use this next week? Why or why not?" Write the answers down verbatim.
- Fix the blockers between testers, rebuild, and send the new APK. With Member's unlimited builds this is a two-minute cycle.
What to measure in the first two weeks
- Activation - of the people who installed, how many completed the core action once? Below half usually means the flow is confusing, not that the idea is wrong.
- Repeat - how many did it a second time without being asked? This is the number investors and your own gut care about.
- Time to core action - minutes from opening the app to booking. Ask the agent to log it to the founder dashboard.
- Drop-off screen - where people stop. One event per screen, counted on the dashboard, is enough.
- Verbatim quotes - the sentences testers said. They become your landing page copy and your next three features.
A dashboard with these five numbers is one prompt: "Add to the founder dashboard: installs (accounts created), users who booked at least once, users who booked twice or more, median minutes from first open to first booking, and a count of the last screen seen before users left." That is your MVP metrics stack, and it costs nothing extra.
When to export the source and hire
Do not hire on Sunday night. Hire when the usage says the idea is real and the next step is something the builder is not the right tool for. The signals:
- Repeat usage is there and you need payments, payouts and receipts at a level a payment provider's hosted checkout does not cover.
- You need integrations with systems the agent cannot reach from a prompt: a specific point-of-sale, a legacy CRM, a hardware device.
- Scale or performance - thousands of concurrent users, heavy media, real-time features across many rooms.
- iOS on the App Store - export the React Native source and have it built with Xcode or a cloud Mac service; the iPhone guide lays out the routes.
- An investor or a partner asks to see the code. Export gives you a complete project as a zip that any developer can open, plus your usage data next to it.
Until then, keep building in the builder: every typed request creates a checkpoint, the signed store build is one click when you are ready for Google Play, and the timeline guide shows how the store's own waits fit around your calendar. What it costs: the free plan covers the weekend for most MVPs; the $1.49 Member plan is worth it on Sunday for unlimited builds and the export you will eventually want. Apps with sign-in, roles and a founder dashboard are the standard MVP shape, and the login and dashboard apps in the gallery show what a clean version of it looks like before you write your own five bullet points.
Frequently asked
What should an MVP app include?
One core action that delivers the value you are testing, sign-in so you can tell users apart, and a way to see whether people come back. Everything else, including payments, a marketing site and an iPhone version, can wait until the core action is used.
Do I need a technical co-founder to build an MVP with AI?
Not for the first version. You describe the app in plain English, the agent writes React Native code and a hosted backend, and you test on a real phone. A developer becomes valuable when the MVP has users and needs custom infrastructure or scale.
How do I get the MVP onto testers' phones?
Download the test APK from the preview header and send the file or link to your testers; it installs on any Android phone once they allow installs from the browser. For iPhone testers, ask for a website version of the same app that shares the backend.
How much does a weekend MVP cost?
The free plan covers building a small MVP; a free APK download is included; unlimited builds for testers are $1.49 on the Member plan. The $1.49 Member plan removes the build cap and adds source export, which you will want for investors or a developer. Google Play's $25 fee applies only if you publish to the store.
When should I export the source and hire?
When real users are doing the core action repeatedly and you need something the builder does not do: custom infrastructure, complex integrations, an iOS release, or performance at scale. Export the complete project on a paid plan and hand it to a developer with the usage data.
Build it now — free to start
Type the idea, sign in, and the agent builds it in your browser. Change anything by describing it.
Open the builder ›