A POC asks one question: can I make it?
A prototype asks if people want it. A first product asks if they pay and keep using it. A POC asks if you can make it. Only you and your team see it, only the hard step is real, and you are done when your number says yes or no.
Find the one block that holds it all up
Every idea is a tower. Pull out the logo, the login or the colours and it still stands. Pull out one block and it falls. That block is your POC. At X, Google's moonshot lab, they say: train the monkey first, the pedestal is easy. Astro Teller tells his teams to "spend zero time thinking about the pedestal".
Start with what you have
Thomas would ask: who are you, what do you know, who do you know? Build your POC with that. Pick a test you can afford to lose, an afternoon and not your savings. And if you can't make the hard step yourself, a partner who says "we can do that" is your proof. That is effectuation, from Saras Sarasvathy's research on expert entrepreneurs.
Write your number down before you test
Decide what counts as proof first, like "8 of 10 right" or "under one second". If you pick it afterwards, every test looks like a win. Large US heart trials show this: before 2000, 57% reported a win. Once researchers had to register their goal in advance, only 8% did (Kaplan and Irvin, 2015).
Build that step and fake the rest
No design, no login, no ratings. Only the hard step is real. In the live demo we tested if Supplist can read real prices at XXL Nutrition, a site that blocks automated price checks. Everything around the price check was left out. Fake the rest, but never fake the step you test.
No code is fine
Pieter Levels started Nomad List as a public Google spreadsheet: can strangers fill in the data for hundreds of cities? It quickly filled up, and the site came five weeks later. You can also act the step out by hand, for real, with a timer: make twenty mocktails in an hour, or call three carriers.
Give it an afternoon
Levels gives a first version one month. Marc Lou caps every build at 14 days. Your POC gets an afternoon. If the step is too big for that, finding the smaller step is your result.
Trust your number, not your feeling
In a 2025 study, experienced developers felt 20% faster with AI and were 19% slower. In the 2026 follow-up they were 4 to 18% faster, and METR calls that weak evidence. Either way your feeling is not a measurement. AI builds, you judge. That is why the number comes first.
A no is a result
Levels counts 4 hits in more than 70 projects. My archive holds 55 folders, and most of them stopped. X pays a bonus to teams that end their own project. A POC that says no this month saves you a product that fails next year.
Bring your number
Run your test for real this week and bring the number to the standup on Friday 30 October. That week we take the other question: do they want it?

