01 / 08
Most founders building with AI right now are making the same mistake: they're letting the technology define the product. They find a capability — a model that summarizes, generates, predicts, or classifies — and they build a surface around it. The result is a product that demonstrates AI rather than solves a problem. Users bounce. Founders are confused. The model was impressive. What went wrong?
What went wrong is that the product thinking came second. This isn't new — it happens with every platform shift. It happened with mobile, with voice, with blockchain. The pattern is consistent: a genuinely powerful new capability arrives, and a significant portion of builders spend the first few years building products about the capability rather than products that use it to do something people actually need. AI is running the same cycle, just faster and louder.
This is an example of a quote
The fundamentals didn't change. Your checklist did.
Good product design has always required the same things: a clear user, a real problem, a reason this solution is better than the alternatives, and a way to know if it's working. AI doesn't retire any of those requirements. What it does is change the cost structure of certain solutions, expand what's technically feasible, and — critically — raise the bar for what counts as a meaningful improvement. If your AI feature saves someone thirty seconds on a task they do twice a month, you haven't built a product. You've built a demo.
The more useful question to ask isn't "what can we do with AI?" It's "what has always been true about this user's situation that we couldn't previously address?" AI often unlocks solutions to problems that were real but previously too expensive or complex to solve well. That's where the product opportunity lives — in the problem, not the model. Founders who start from user reality and work toward the technology tend to build things that stick. Founders who start from the technology and work toward a use case tend to build things that demo well and churn fast.
Where AI actually changes product design
That said, AI does change some things meaningfully, and it's worth being precise about what they are. First, it changes the interaction model. When a product can respond to natural language, adapt to context, or generate output rather than retrieve it, the design of that interaction requires different thinking. The old patterns — forms, filters, rigid navigation — often don't fit. You're designing a conversation or a collaboration, not a transaction. That requires more attention to expectation-setting, error states, and what happens when the output is wrong or unexpected. Users don't yet have strong mental models for AI behavior, which means your product has to do more work to build trust incrementally.
Second, AI changes the feedback loop between product and user in ways that affect how you measure success. Traditional product metrics assume relatively deterministic behavior: a user clicks a button, a thing happens, you measure the outcome. AI-powered products introduce variance. Two users asking the same question may get different answers. The same user asking the same question twice may get different answers. This isn't necessarily a problem — it can be a feature — but it means your quality bar and your success metrics need to be designed with that variance in mind. Measuring whether an AI feature is working requires more rigor, not less, precisely because the outputs are harder to audit at scale.
The senior judgment question
The most useful thing a founder can do right now is apply the same skepticism to their AI product decisions that a good investor applies to a pitch. Not cynicism — skepticism. Ask whether the problem is real before you ask whether the model is capable. Ask whether users would notice if the AI were removed, or whether they'd barely register the difference. Ask whether you're adding AI because it genuinely makes the product better, or because it makes the product feel more fundable. Those are different motivations, and they produce different products.
AI is a legitimate capability shift. It will change how a lot of software gets built, and it will enable products that weren't possible before. But it won't change what makes a product worth building: a real user, a real problem, and a solution that's meaningfully better than what they have now. The founders who hold onto that through the noise — who use AI as a tool rather than a thesis — are the ones who'll have something durable when the hype settles.