This week an engineer with a large following proposed a rule for AI-written code: you should be able to defend every decision your system makes, even if a machine wrote it. The same day, a CFO told the story of an AI image winning a state fair, and drew the obvious conclusion, that when creation gets cheap, judgment gets expensive. In founder forums, the same argument ran under a different headline: now that anyone can build an app, what is the moat?
All three conversations are circling one uncomfortable fact. The MVP, the sacred first milestone of startup life, has quietly stopped proving anything.
Why did an MVP used to feel like proof?
For twenty years, building was the filter. Shipping a working product required either engineering skill, money to hire it, or the persuasion to recruit a technical cofounder. Most people had none of the three, so most ideas died unbuilt. An MVP that existed at all carried information: someone cared enough, and was capable enough, to make it real.
Investors read it that way too. A prototype separated the talkers from the builders. That was never demand evidence, but it was scarcity evidence, and scarcity is easy to mistake for signal.
What changed when AI made building cheap?
The filter is gone. A working web app is now a weekend of prompting. The supply of shipped products has exploded, and every one of them looks polished, because polish is exactly what the tools are good at.
When everyone can produce the artifact, the artifact stops carrying information. This is not a moral judgment about AI tools. It is just what happens to any signal when it becomes free to emit. A shipped MVP in 2026 tells the market roughly what a business card told it in 1995: this person exists.
What does an MVP actually validate?
Here is the part that was true even before AI: an MVP never validated an idea. It validated feasibility, which is a question about you. Demand is a question about other people, and no amount of building answers it.
The confusion survives because building feels like progress. Six months of engineering produced visible artifacts, commits, screenshots, a launch date. Asking strangers whether they would pay produces mostly silence and the occasional bruise. Founders chose the activity that felt better, and called it validation.
AI removed the alibi. When the build takes a weekend, you can no longer spend half a year hiding inside it. The judging arrives immediately, and it turns out the judging was always the hard part.
What counts as proof of demand now?
Costly actions by strangers. That is the whole list. A cost can be money, which is the clearest version. It can be time, when someone returns to the product for a second week without being nudged. It can be reputation, when a buyer pushes your tool inside their company. What all real proof shares is that the person giving it pays something to give it, which is exactly why compliments do not qualify. Agreement is free.
The stranger part matters as much as the cost part. Friends and family will spend money on you, not on the idea. Their yes is a gift, and gifts are not data.
Should you still build an MVP?
Yes, and faster than ever. Cheap building is genuinely good news. The mistake is not building, it is treating the build as the finish line of validation instead of the starting gun. The MVP is an instrument. Its only job is to put something in front of people who owe you nothing and record what they do, not what they say.
So build the weekend version. Then spend the six months you just saved on the question that was always the real one: will a stranger pay for this? Everything else is a business card.
Key takeaways
- An MVP validates feasibility, which is a question about you. Demand is a question about other people, and building never answered it.
- AI made shipped products abundant, and abundant signals stop carrying information.
- Proof of demand is a costly action by a stranger: payment, repeated unprompted use, or reputation spent on your product.
- Compliments, survey answers, and sign-ups from friends are free to give, so they prove nothing.
- Keep building MVPs, but treat the build as the starting gun of validation, not the finish line.
FAQ
Does building an MVP validate a startup idea?
Not by itself. An MVP is a tool for collecting evidence, not evidence in itself. The validation happens when strangers use it repeatedly, pay for it, or change their behavior because of it.
What should I validate before building anything?
Whether the problem is painful enough that real people already spend money, time, or workarounds on it. What people currently do about a problem is stronger evidence than anything they say about your solution.
Is an MVP still worth building in the AI era?
Yes, because it is cheaper than ever. What changed is its meaning. A weekend build is a fine instrument for testing demand, but it is no longer an achievement or a signal on its own.
What counts as real proof of demand?
Costly actions by strangers. Payment is the clearest one. Repeated unprompted use, a stranger asking about price before you raise it, or someone abandoning their current workaround all count.
Why do investors care less about prototypes now?
Because AI collapsed the cost of producing one. When everyone in the room can show a working demo, the demo stops separating anyone. What separates founders now is evidence that people with no reason to be polite want the product.