I wrote recently that in the age of AI, where everyone has access to tools to create software, speed is not strategy. If code is becoming a commodity, I argued, then the advantage no longer lies in how fast you can build. It lies in how deeply you understand your customer and their problems. That is still true. But there is another mistake founders can make with AI, and in some ways it is even more important.

Building Is Cheap. Insight Is Not.

Once building is cheap, the temptation is to treat understanding as cheap too. To assume that if AI can write the code, it can also do the discovery. It cannot. Writing the code and understanding the customer are two different jobs, and AI is only good at one of them.

While everyone is racing to use AI to build, the bigger opportunity is to use it to learn: to run sharper, faster customer discovery and pull real insight out of it. Building used to be the scarce resource. It is why everyone went looking for a technical cofounder. That ship has sailed. What is scarce now, and what actually decides whether a company lives or dies, is real insight into a real problem and the people who have it. That insight does not come from a prompt. It comes from the work of customer discovery.

And that work is stubbornly human. It means having a point of view worth testing. It means the hustle to get the interviews, and the discipline to actually conduct them. It means asking the right questions, listening for the signal in an offhand comment, and knowing when to probe: to ask the follow up, and then to ask why, and why again, until you reach the root of the problem. No tool does that for you. A founder who skips it does not get insight from AI. They get a confident summary of nothing. 

This is why I am wary when AI is sold as a shortcut to strategy. The prompt for this piece was a pitch that landed in my inbox, from a startup promising to turn a one line idea into a full go to market strategy. Type in an idea, get back a market, a business model, and a roadmap, polished, confident and persuasive. It is tempting to take that as the answer. It is not. A strategy you did not earn through customer discovery is just a guess in a nicer package. Do not put your strategy on autopilot.

The One Part You Cannot Outsource

This is why customer discovery is the one part you cannot outsource. You can hand AI the prompt to write the code. You can hand it an elevator pitch and have it tighten it up. You cannot hand it the judgment about what a customer actually meant, which problem is worth solving, or whether the enthusiasm in an interview was real or just polite. That judgment is earned by doing the work, and it stays with the founder.

We watched this play out with one of our founders earlier this year. Yannick Bierens is a builder by instinct. His first startup, a social app for discovering restaurants, grew to 250,000 users sharing a million places. He came into our J-Term Startup Sprint building Visibly, a tool to visualize codebases, born from his own frustration losing track of AI generated code. Before the Sprint he believed he had validated the idea after talking to five developers. During the Sprint and in the weeks that followed, he did roughly fifty interviews. By his own account, the first twenty taught him how to have a real discovery conversation at all. It took another twenty for him to see that the problem he was solving, while real, was not painful enough to build a company on. He walked away. As he put it: “if I had not learned how to do customer discovery, I would probably still be working on it and failing.”

What he did next is the more telling part. Rather than reaching for another idea, he treated discovery itself as his asset. He picked a large underserved vertical, insurance, and committed to one hundred customer conversations in five weeks. About thirty in, brokers kept volunteering the same unprompted problem: young brokers struggle so hard to land their first clients that most wash out within a few years. That earned insight is the foundation of his new startup, Binder, and this time he is proving he can deliver value by hand before he builds a thing. That is what it looks like when the founder, not the machine, holds the judgment. 

It took fifty conversations to walk away from Visibly, and no small amount of work to make sense of them. Watching him do it changed how we think about teaching it.

Where AI Earns Its Place

So where does AI come in? Exactly where the human limits are. A founder who runs dozens of interviews ends up sitting on troves of messy, unstructured input: transcripts, notes, half-remembered conversations, contradictions. No one can hold all of that in their head. This is where AI earns its place. Its first job is to keep the founder honest. It holds them to what was actually said, rather than what they think they heard, or wish they had. It can also organize that record, surface patterns across it, and sharpen the questions worth asking next. But be clear about what that is. A pattern is not an insight. AI can tell you that a dozen customers used the word frustrating. Only the founder who sat in those conversations can tell you whether the frustration is deep enough to build a company on. If you catch yourself treating the patterns as the conclusion, you have not multiplied your judgment. You have outsourced it. It does not replace customer discovery. It multiplies what the founder can learn from it. 

Used well, it does something no coach or mentor can do at scale. It sits in on every interview. It tells a founder not just what they heard, but what they failed to ask, where a question was leading, where a customer was describing someone else’s behavior rather than their own. Most founders get that kind of feedback on one interview in twenty, if they are lucky enough to have a mentor in the room. The point is not that the machine knows the answer. It is that it can tell you when you have not yet earned one.

A Multiplier, Not a Substitute

That is the force multiplier. And the word multiplier matters. AI amplifies what is already there. A founder who is doing the hard work of customer discovery gets sharper and faster with these tools, and, used well, more honest. We all suffer from “entrepreneur happy ears”, the pull to hear what we want to hear. The right system pushes back. It can flag where you were pitching and selling instead of testing the problem and the customer, and where you counted enthusiasm that was not really there. A founder who is not doing the work gets none of that. They stay exactly where they were, just with better-looking slides and a product no one likely wants. 

I have seen this movie before. Too many times. The tell is in the language. A founder comes back from a week of conversations and reports that a customer “likes our solution”. Likes it. Not addresses an urgent problem for them. Not uses it, not pays for it, not abandoned the workaround they built in a spreadsheet three years ago to switch to it. Liked it. Founders who sell instead of doing discovery confuse curiosity or politeness with genuine interest, and those are not the same thing. Most people will be curious to learn about your product for twenty minutes. Almost no one will change how they work unless it solves an urgent problem for them. The updates fill with pipeline and outreach, the slides get better, and from the outside the team looks like it is moving faster than everyone else. It is not moving at all. None of that is new, and none of it is AI's fault. But AI makes it faster, and it makes it prettier. The tools can not turn a weak founder into a great one. They make a good one greater.

This is why I have long argued that we do not bet on ideas, or even on technologies. We bet on founders. As I have written before, there are many startups with no patents, but there are zero startups with no founders. AI does not change that. If anything, it raises the stakes. When the tools are this good and this widely available, the idea is no longer the differentiator, and neither is the code. The founder is. What separates the ones who break through is the judgment to know what they are looking at, the discipline to keep listening when the easy answer is right there, and the insight to see what others miss. AI can sharpen all of that. It cannot manufacture it. For the founder who is already chasing insight, AI is a force multiplier. For the one who is not, it is just a faster way to fool yourself.

Which raises the real question, and the one my colleagues have been working on. If AI is going to multiply discovery rather than replace it, what does that actually look like in practice? How do you put it to work on the messy evidence of customer discovery so that founders keep custody of the meaning while the machine helps them see it more clearly? That is what we have been building toward, and it is what they will write about next.