How to Pick a Startup Idea, Per Y Combinator

There's a guy I think about whenever I cover this particular flavor of founder paralysis. He ran a specialty food importing business out of a warehouse near the rail yards — good product, genuine relationships with suppliers across three countries, a customer base that kept reordering. For two years, he kept a whiteboard in the back with six other business ideas he was "evaluating." A catering arm. A retail storefront. A wholesale licensing play. An e-commerce channel. Every quarter he'd shuffle the list, do a little research, talk to a few people, and put it back on the whiteboard. The core business grew modestly. The whiteboard stayed full. He wasn't failing, exactly. He just wasn't going anywhere.

I think about that whiteboard when I watch Jon Xu, a general partner at Y Combinator, deliver his latest Startup School episode on how to pick an idea and commit to it. The advice is aimed squarely at software founders circling the drain of pre-launch indecision. But the thing that makes Xu's framework interesting — and genuinely worth your time whether or not you've ever heard of a seed round — is that the underlying diagnosis is dead accurate about something much older than startups.

"The worst failure mode isn't being wrong," Xu says. "It's not making a decision, spinning your wheels, dabbling between ideas, and never going deep enough on any one of them to learn anything."

That's the whiteboard problem. And it lives on Main Street as much as it lives in San Francisco.

---

The paralysis has a shape

Xu identifies two versions of overthinking that keep founders stuck. The first is hunting for the perfect idea before committing — understandable, he says, but ultimately impossible, because you can only figure out what works by making contact with reality. The second is convincing yourself you're not the right person for the idea — that you lack the credentials, the background, the years in the industry.

He's right on both counts, and I'd argue the second one is the more pernicious in a small business context. The food importer I mentioned wasn't waiting for the perfect idea. He had plenty of ideas. What he was really doing was waiting for permission — a clear enough signal that one direction was safe before he walked away from the others.

Xu's prescription is blunt: "Burn the other boats." Pick one idea, explicitly foreclose the others, tell customers you've pivoted if you need to, and work with single-minded focus on what you've chosen. He illustrates this with GovDash, a YC-backed company that helps customers win government contracts. According to Xu, the team pivoted multiple times before landing on that idea, changing their company name and contact details with each pivot — at one point Xu says he lost track of how to reach them because they'd changed their email addresses again. Their series B raise, which Xu mentions in the video, hasn't been independently verified here, but the underlying point about commitment is his, not mine.

The burn-the-boats framing is vivid and I mean this sincerely: it's correct. Splitting attention between options doesn't give you information about either one. It gives you noise about both.

---

But let's be honest about what burning boats costs

You can't always burn the other boats when you need them for rent.

This is where Xu's framework, built for founders who either have runway or are about to apply for YC funding, brushes up against a different reality. The bootstrapper who's building a service business on nights and weekends while keeping a day job isn't hedging because they lack conviction. They're managing actual risk with actual money. The optionality isn't a psychological failure. It's a survival structure.

Xu isn't wrong that divided focus produces bad signal. He's just describing a situation where the cost of being wrong has already been underwritten. When YC is in the picture, or when you've got investor capital in the bank, you can afford to be totally wrong fast and call it learning. When the business is also your income, being totally wrong fast is just called being broke.

That's not a knock on Xu's advice. It's a translation note. The principle — commit deeply, gather real data, let the work surface the better idea underneath — holds. The timeline and the acceptable risk level are different.

---

"Could you run their business?" is the right question

Here's the part of the video that deserves to travel well beyond startup world, and I mean that without qualification.

Xu's benchmark for whether a founder has gone deep enough on an idea is a single test: could you actually run your customer's business? Not "have you talked to twenty of them." Could you be dropped into their operation tomorrow and keep it moving?

He uses the example of building voice agents for cleaning services. The question isn't whether you understand the software problem. The question is: do you know what their daily crises are? Do you know whether answering the phone is a top-five problem or a top-twenty problem? Do you know exactly how much revenue walks out the door when a call goes unanswered, and what a cleaning business owner would actually pay to stop losing it?

What Xu is calling customer depth here is something any good sales rep on your block already understands instinctively — it's the difference between the vendor who emails you a catalog and the one who shows up in January, before you've called, because they know your slow season and they've already thought about what you'll need in March. It's the accountant who's been with you since year two and sends you a one-line note in November because she noticed your receivables are running long. Those people know your business well enough to run it. That knowledge is the relationship. That's why they keep the account.

Xu frames this as a startup validation technique. Fine. But the underlying principle — that deep familiarity with how another person's operation actually works is where real value lives — is older than venture capital and more durable than any particular market.

"Getting to this level will involve lots of conversations with customers," Xu says, "and sometimes even literally doing the job yourself."

The people on my beat who build businesses that last usually figured that out the hard way, without a framework. They just called it paying attention.

---

The AI layer, and what it actually means for smaller operators

Xu spends a meaningful portion of the video on what makes a good idea specifically in the current moment, and here the advice gets more specifically targeted at software founders. His argument is that the best opportunities right now sit at the edge of what AI models can currently do — products that barely work today but will clearly improve as the underlying technology gets better. He also argues that the real prize isn't building software tools for existing industries. It's owning the outcome directly. Don't build software for insurers, he says. Be the insurer.

His example is Corgi Insurance, a company from YC's summer 2024 batch that, according to Xu, set out to own the entire commercial insurance stack rather than just a piece of it — and took what Xu describes as the unprecedented step of acquiring an insurance carrier during their YC batch to make it happen. That claim is Xu's characterization; insurance carrier acquisitions are heavily regulated and capital-intensive, and I can't independently verify the specifics from this piece. But the underlying strategic logic he's articulating — own the outcome, not just the tooling — is worth taking seriously regardless.

For the small business owner or bootstrapped founder, this translates to something more familiar: stop competing on features, compete on results. The service business that guarantees the outcome rather than just providing the labor. The software tool that comes with implementation and accountability rather than just a license key. The more you own the outcome, the harder you are to replace.

---

What to do when the idea doesn't work

Xu closes with a reframe that I think is genuinely useful and gets undersold in most startup advice: going deep on an idea that ultimately fails is not a waste of time. It's the process by which you find the better idea underneath.

"When most founders begin, they're solving surface-level pain points," he says. "The real opportunities are almost always the deeper structural problems."

This is true. It's true in tech and it's true in every other kind of business. The person who went deep on the wrong idea and came out the other side with unambiguous customer data and real understanding of the market's structural problems is in a much stronger position than the person who spent the same time shuffling a whiteboard.

Pick one direction. Walk fast. The destination you couldn't see from the starting point might be the right one.

---

— Dorothy "Dot" Williams, Small Business & Entrepreneurship Correspondent, Buzzrag