How do I build an MVP for my startup, step by step?
Write down the one assumption that kills the idea if it is wrong, then build the cheapest thing that tests it and get it to real users within weeks. Step by step: talk to customers until you can describe their problem in their words; pick the MVP type: a landing page that takes sign-ups or pre-orders, a concierge version you deliver by hand, or a working product; build it with the fastest tool you can use, such as Carrd ($19 a year) for a page, an AI builder like Lovable (free to start) for an app, or a developer if the product truly needs one; cut every feature that does not test the assumption; launch to the people you interviewed; then watch whether they come back and pay, and decide what to change. The rule: if it will take more than a few weeks, the scope is too big.
The single most-cited practical talk on scoping a first product, from the person who ran YC's accelerator. It cuts through the theory and shows what an MVP actually looks like using Airbnb, Twitch, and Stripe.
Michael Seibel, who ran YC, makes the blunt case for shipping something almost embarrassingly simple and learning from real users instead of chasing a perfect first version. His line that if it takes more than a month to build, it is not an MVP is a practical yardstick for anyone stuck polishing. Watch it when the urge to add one more thing before launch takes over.
A longer office hours style session where two YC partners answer real founder questions about scoping and launching a first version. Hearing them push back on founders who want to build too much helps you calibrate your own MVP. Watch it when you are stuck deciding what to cut before you launch.
The single best thing ever written on customer conversations. It teaches you to ask about the customer's life and past behaviour, not your idea, so you can't be lied to. If a founder reads one thing before talking to a single customer, it's this.
The two ways to fake a product while you test real demand. Do the work by hand in the open (concierge), or hide the manual work behind a front so users think it is automated (Wizard of Oz, the way Zappos started). Clear examples of when each one fits.
The canonical fake-door test, told by the founder who ran it. Buffer started as a two-page site: one page pitched the tool, a pricing button led to an email capture, and every signup got a personal note. Real demand, real conversations, a paying customer in weeks, and not a line of product code.
The permission slip to recruit users by hand, do things manually, and deliver 'insanely great' experiences to your first few customers. The cheapest, most honest way to validate demand is to go get it one person at a time.
The fastest way to stand up a fake-door landing page. Build a one-page site in an afternoon, add an email capture, and measure whether anyone actually tries to sign up for what you are describing. Free to start, no code.
Lovable is an AI app builder: you describe the website or app in plain words and it builds, hosts and publishes it. Its plans page shows the free plan with 5 build credits a day (up to 30 a month), and paid plans from $25 a month for 100 credits ($21 a month billed yearly).
