How To Pick A Startup Idea

Why does it matter? Because the right startup idea is hiding underneath the wrong one you haven't committed to yet

Most founders treat idea-selection like something to get right before the real work starts. John from YC inverts this entirely: you cannot evaluate an idea from the outside, only from inside it. The only way to find what you should build is to go deep on one thing — not several — until the real opportunity surfaces from underneath.

• In the AI era, the cost of software is collapsing toward zero; the moat is licenses, trust, and owning the outcome — not tooling for an industry
• "Going deep" means being able to run your customer's business if dropped into it tomorrow — not logging 20 discovery calls
• Running two ideas in parallel destroys signal for both, leaving you unable to distinguish a good idea from a bad one
• The worst failure mode isn't committing to the wrong idea — it's never going deep enough on any single one to learn what's actually underneath

In the AI era, building software for an industry is a losing position — just become the industry

"The cost of producing software is going to zero." John drops this early, and it rewrites the logic of every B2B startup built on selling tooling to an industry vertical.

If software costs nothing to produce, software-for-X companies are running toward commodity pricing. What holds value are the things a better model can't replicate next quarter: customer trust, regulatory licenses, and economic ownership of the outcome. The instruction is blunt: "Don't build software for insurance companies. Just be the insurer."

Corgi Insurance — from YC's Summer 2024 batch — is the case study. They rejected the position of tech-enabled broker and rejected managing general agent status too. Both own only "part of the solution." Instead, they aimed at the entire commercial insurance stack — underwriting, customer service, everything — and took the unprecedented step of acquiring an actual insurance carrier during their YC batch. The result: they can underwrite any insurance line in any vertical with a fraction of traditional carriers' headcount, offer better pricing, faster turnaround, and own all the economics.

The question to pressure-test your idea: are you building a tool for an industry, or are you becoming the industry? The software layer is table stakes now. The moat is everything that surrounds it.

Twenty customer conversations isn't going deep — the real bar is whether you could run their business tomorrow

Drop someone into a cleaning business tomorrow. Could they run it? That's John's actual benchmark for customer knowledge — not whether you've talked to 20 owners.

It means knowing what their daily crises look like. Whether answering the phone ranks in their top five problems. Exactly how much business they lose when a call goes unanswered, and what they would actually pay to never miss one again. Not roughly — with high confidence.

Getting there often means doing the job yourself, not just shadowing it. Build in a tight loop: deep customer understanding → product → deeper understanding → better product. Real customers using what you've built generates concrete data that no amount of desk research produces. The goal isn't to talk to hundreds of people before writing code — it's to run both in parallel, each sharpening the other.

The self-test: could you teach a class on this problem? Are you one of the most informed people in the world on it? That standard sounds extreme. It's also what separates products that solve visible surface pain from the ones customers would actually pay to have fixed.

Juggling two startup ideas doesn't split the risk — it poisons the data for both

Bad signal is worse than no signal. Split your time between ideas rather than going deep on one, and you can't generate reliable data on whether anything is working. With ambiguous data, you fall into a trap with two jaws: kill a good idea too early, or convince yourself a bad one deserves more time. Both errors trace back to the same root cause.

The fix is sequential depth. Pick one, then burn the other boats — literally. Stop working on the others. Tell any customers you've pivoted. Change the company name, the email addresses, the social media, the internal narrative about why you're building at all. Close the other options in a way that feels irreversible, not just paused.

John's bar: going deep should feel like wearing a new skin. "You should become an almost unrecognizable version of yourself." GovDash — which helps customers win government contracts — pivoted at least five times before landing their idea. Each time, they changed names, emails, and their entire company story. John says he once lost track of how to reach them because they changed contact information so completely with every pivot. Their fifth idea worked so well they could barely keep up with demand. They've since raised a Series B.

Going deep on a failed idea is the mechanism — the real company is almost always buried underneath

The fear that going deep on the wrong idea is wasted time is exactly the fear that prevents founders from finding what they should actually build.

"Going deep isn't primarily a process for validating the idea you started with. It's a way to find the better idea underneath." Founders begin by solving surface-level pain — problems visible from the outside. Structural opportunities live deeper, and you only see them after being inside long enough to notice the bottlenecks, the gaps, the dev tools nobody's built. One of those might be the actual company.

This is especially true at the frontier of what models can do today. Your product might barely work on current models but will clearly improve as they get better — and the specific bottleneck you hit while building it could turn out to be the real opportunity.

A failed deep dive leaves you with something real: unambiguous data on whether there's a hair-on-fire problem in the space, genuine conviction for any pivot, and a much cleaner view of where the structural gaps are. That starting position is dramatically stronger than someone who spent the same months switching between ideas without going deep on any.

Don't fear being wrong. Fear not going deep enough to see what's underneath.

A wildly ambitious startup costs roughly the same as a modest one — so the math only points one direction

"The cost of pursuing a wildly ambitious startup idea and the cost of pursuing a modest one are roughly the same. They're both extremely hard. They both place extreme demands on your time."

If you're going to suffer either way, aim at the version that rewrites a sector of the economy if it works — the version that protects you from competitors, attracts the best talent, and builds a moat worth defending. Practically: the most regulated industries (legal, healthcare, financial services), going head-to-head with a $10 billion legacy SaaS incumbent, or building hard tech like robotics for space assembly.

Modest ideas carry all the costs of startups with none of the upside asymmetry. Size up deliberately.

The certainty you want before committing can only come from the commitment you're avoiding

"It's impossible to figure out the perfect idea in the abstract. You can only figure out what you should be working on by making contact with reality and getting feedback from customers."

The early-stage temptation is cautious steps in every direction — sample a little here, a little there, stay close to home. That approach generates almost no information. What generates information is committing to one direction and walking fast. You're not guaranteed to arrive in the right place. But you produce far more signal per unit of time, and when you're moving, you might end up somewhere better than anything visible from your starting point.

Stop analyzing. Pick any idea that clears a basic bar and go.

Second-time founders weaponize 'founder-market fit' as a reason to never start

Blake Scholl spent his early career in ad tech at Amazon and Groupon. Then he decided to commercialize supersonic flight. The domain mismatch looked absurd to most observers. Boom Supersonic is now the proof that it wasn't.

Founder-market fit is real — a non-technical founder probably won't generate a killer dev tool idea. But second-time founders especially "weaponize this line against themselves," convincing themselves they need a decade of domain experience before they're allowed to start. The truth: if you pick an idea you're genuinely curious about and go extremely deep, "it's often possible to develop extraordinary knowledge in a short amount of time."

Domain experience is less a prerequisite than an output. Don't let its absence be permission to delay.

The founders who win in AI won't have the best initial idea — they'll generate the best information fastest

As AI capabilities compound quarterly and software costs collapse toward zero, initial idea quality matters less than information velocity. The winners will be founders who committed early, went deep, and stayed inside a problem long enough for the structural opportunity to become visible from the inside.

The right idea doesn't exist until you're deep enough in the wrong one to see it.

---

Topics: startups, ideation, founder advice, Y Combinator, AI era strategy, customer discovery, founder-market fit, vertical integration, commitment, product-market fit