How to pick a startup idea

Stop trying to find the perfect idea

A lot of founders have several ideas and can’t decide between them. Often they are working on a few at once, and usually they tell themselves they are waiting to find the best one before fully committing. The trouble is that it is nearly impossible to make real progress on a startup without committing to a single idea, so this is worth saying plainly, and it is really a rubric for how to stop overthinking, pick something, commit, and then find out fast whether it works. The two most common ways founders overthink are both worth naming.

The first is believing a perfect idea exists to be found. It is an understandable impulse — startups are hard, so surely you should figure out the best one before you commit. But you cannot figure out the perfect idea in the abstract. You only learn what you should be building by making contact with reality and getting feedback from customers. The second is asking whether you are the perfect founder for this. Founder-market fit is real — a non-technical founder probably isn’t the one to nail a killer devtools idea — but founders, especially second-timers, weaponize this against themselves and decide they need a decade of domain experience before they can start. They don’t. Pick something you are curious about, go extremely deep, and above all talk to customers, and you can build extraordinary knowledge fast. Blake Scholl spent his early career on ad tech at Amazon and Groupon before deciding to commercialize supersonic flight; plenty of people thought he was crazy, and now Boom is a billion-dollar company. Don’t let the question of whether you are allowed to work on something stop you from starting.

Commit to one, and burn the other boats

Once you have stopped overthinking, commit to a single idea. Founders who juggle several at once usually believe that is how you discover which one works. The most serious problem with that is bad data: if you don’t go deep on any one idea, you get weak signal on all of them, and weak signal is worse than none — it lets you talk yourself out of a good idea or into a bad one.

So go deep, and the first move is to burn the other boats. Explicitly foreclose your other options — stop working on them, tell any customers you have pivoted, and work with single-minded focus on the one you chose. Done right, going deep should feel like wearing a new skin: you become an almost unrecognizable version of yourself, sometimes changing the company name, the emails, the website, even your internal story about why you are building this at all. I worked with a startup, GovDash, that helps customers win government contracts; they pivoted at least five times before landing there, changing their name and mission each time — at one point I couldn’t reach them because they had changed their email addresses again. By becoming genuine experts in government procurement, their fifth idea worked well enough that they could barely keep up with demand, and they recently raised a Series B to meet it.

The bar: could you run your customer’s business?

Once you have committed and gone deep, how do you know you are doing it well? The high watermark I use is this: could you actually run your customer’s business? Say you are building voice customer-service agents for cleaning services. The question is not whether you have talked to twenty owners. It is whether, dropped into a cleaning business tomorrow, you would know how to run it — what their daily crises are, whether answering the phone is even a top-five problem, how much business they lose when a call goes unanswered, and what they would pay to never lose another one. You need to answer questions like those with real confidence. Another way to put it: could you teach a class on the problem you are solving? Are you one of the most informed people in the world on it?

Getting there takes many customer conversations and sometimes literally doing the job yourself. But don’t obsess over talking to hundreds of customers before you write any code. The goal is to do both at once, in a tight loop: deep understanding of the customer, then product delivery, then deeper understanding, then better product. Real customers using a real product generate concrete data that complements your abstract knowledge and tells you whether the thing is actually working.

What a good idea looks like in the AI era

Pull from customers is the most obvious signal, but as you go deep, watch for three qualities that mark a strong idea right now. The first is that it sits at the edge of what models can do today. Your product might barely work on today’s frontier models but clearly get better as they improve. Understand intimately the bottlenecks holding back its performance, because if a particular bottleneck doesn’t clear the way you hoped, solving it might become the company. This is Paul Graham’s “live in the future and build what’s missing,” restated for the model curve.

The second is that it should verticalize — ultimately sell an outcome, not software for X. As the cost of producing software goes to zero, the durable value moves to customer trust, licenses, regulatory permission, and outcome ownership. So don’t build software for insurance companies; be the insurer. Corgi Insurance, from YC’s Summer 24 batch, refused to settle for being a tech-enabled broker or even a managing general agent, because that only owns a slice of the solution. They set out to own everything from underwriting to customer service — the whole commercial insurance stack — and even acquired an insurance carrier during their batch to make it happen. Owning the full stack lets them underwrite any line in any vertical with a fraction of a traditional carrier’s headcount, offer better pricing and faster turnaround, and keep all of the economics.

The third is that it should be the most ambitious version of itself. This seems backwards, but the cost of pursuing a wildly ambitious idea and a modest one is roughly the same — both are extremely hard and both consume your life. So aim at the version that, if it works, rewrites a sector of the economy, because that is also the version that repels competitors, attracts the best talent, and has a moat worth building. That might mean the most regulated industries, taking on a ten-billion-dollar legacy incumbent, or building genuine hard tech.

Even if it fails, you win

Do all of this and have the idea fail, and you are still dramatically better off than where you started. You have unambiguous customer data — you know whether there was a real hair-on-fire problem or whether you talked yourself into one — so any pivot rests on real conviction and a better sense of how to execute.

More importantly, going deep usually hands you the better idea underneath. Most founders start by solving surface-level pain; the real opportunities are almost always the deeper structural problems, and going deep is less a way to validate the idea you started with than a way to find the one hiding below it. This happens especially at the frontier of what models can do — you notice the bottlenecks, the gaps, the devtools nobody has built, and one of those turns out to be the actual company.

So here is the whole thing. Stop hunting for the perfect idea; pick one. Burn the other boats. Learn everything you can about the customer and try to execute for them. In the early fog, where you can see ten feet ahead, the temptation is to take cautious steps in every direction — and that gives you almost no information. Commit to one direction and walk fast. You are not guaranteed to end up in the right place, but you generate far more information per unit of time, and you might arrive somewhere better than you could have seen from the start. The worst failure mode isn’t being wrong; it is never deciding — spinning your wheels, dabbling, never going deep enough on anything to learn from it.

Based on Jon Xu’s YC Startup School talk on picking a startup idea. The rubric is his; this version is my synthesis and wording.

Related reading

• How to get your first ten customers — what “going deep” and “talk to customers” look like as concrete tactics.
• How to build an AI-native services company — verticalizing an idea all the way into owning the outcome.