Rob Fitzpatrick wrote The Mom Test in 2013. The central problem it solves is this: people lie to you, not out of malice, but because they want to be nice. Your mom tells you your idea is great. Your potential customer tells you they'd definitely pay for it. Neither of them means it.
The book's answer was simple - stop asking for opinions and ask about their actual life instead. Past behaviour over future intent. Specifics over hypotheticals. I wrote up my notes on the book separately if you want the details.
Thirteen years ago, that was enough. Now it's not.
I built something for two months that nobody wanted
I run Albion Online Grind as a side project. At the very start, I spent two months building a feature I was convinced the community needed. I shipped it. Almost nobody used it.
The painful part is that I had access to users the entire time. I could have asked. I didn't - not properly. I assumed I understood the problem well enough. I had been in the space for years. I had opinions about what players wanted.
That is exactly the trap the book describes. Conviction is not the same as validation. And the longer you build before testing the assumption, the more it hurts when you find out you were wrong.
What changed in thirteen years
AI has made it cheaper and faster to build things. What used to take months of engineering effort now takes days or weeks. That sounds like a good thing, and it is - but it quietly shifted the bottleneck.
The bottleneck used to be building. Now it's knowing what to build.
And that means the pressure to skip proper customer discovery has never been higher. You can spin up a working prototype in a weekend. The temptation is to ship first and validate later. Or worse - to not validate at all, because moving fast feels like progress.
But speed of building has nothing to do with whether anyone actually wants the thing.
The new failure mode: AI as a fake customer
Here's something that wasn't possible thirteen years ago: you can now ask an AI to roleplay as your target customer and give you feedback on your idea.
I've spent a lot of time doing exactly this. I would prompt the LLM to be a CEO, a CTO, a game designer, a product manager, a potential player - cycling through perspectives, trying to simulate what day zero of a product might look like before building anything. It felt rigorous. It felt like homework.
It's genuinely useful for stress-testing your thinking. But it's also dangerous if you start treating it as validation, which I was doing.
An AI will give you articulate, coherent, seemingly realistic feedback. It won't be rude. It won't be bored. It won't tell you the thing you're building solves a problem they've never actually had. It's the ultimate mom - it wants to be helpful, and helpfulness in this context means telling you what sounds right, not what's true.
The Mom Test rule still applies: you can't validate demand by talking to something that has no skin in the game. An AI has no budget, no actual problem, and no real alternatives it's currently using.
What AI is actually good for in customer discovery
That said, AI has made some parts of the process genuinely better.
Before talking to a real person, you can use AI to anticipate objections, find gaps in your questions, and stress-test your assumptions. You show up sharper.
If you're doing ten or twenty customer interviews, pattern-spotting across transcripts is tedious. AI handles that well - summarising, clustering themes, flagging contradictions.
Cold outreach is easier than it's ever been too. Finding the right people to talk to, drafting the message, following up - all faster now.
None of this replaces the conversation. It makes everything around the conversation more efficient.
The core advice hasn't changed
Fitzpatrick's rules hold exactly as well in 2026 as they did 13 years ago:
- Ask about their life, not your idea
- Specifics in the past, not hypotheticals about the future
- A compliment is not validation
- Real commitment costs something - time, money, reputation
The temptation to use AI to simulate the market instead of talking to it is stronger than ever. The shortcut has always been there. AI just made it more convincing because it gives you something that looks and feels like a customer conversation without requiring an actual customer.
Don't confuse the two.
The thing that still can't be replaced
Real customers say things that surprise you. They use your product wrong. They have a problem adjacent to the one you thought you were solving. They already pay for something you assumed was free. They don't care about the feature you spent the most time on.
That unpredictability is the signal. An AI won't surprise you in that way, because it's drawing on patterns of what a reasonable customer might say, not on the messy reality of what an actual person in that situation actually experiences.
The Mom Test was written to get founders out of their own heads and into the real world. That goal is unchanged. If anything, in an era where you can stay in your head more convincingly than ever before, it's more important than ever to leave it.
Now, whenever I get the chance, I try to jump on a quick call with my users. Not a formal interview, just a conversation. It's the one thing that consistently tells me something I didn't already know.