Pick One Idea and Go Deep — shantum's notes

ELI5/TLDR

Founders waste months comparing startup ideas when they should just pick one, commit fully, and learn fast through customer feedback. Going deep beats sampling. The worst mistake isn’t being wrong—it’s indecision. Pick one direction, walk fast, and you’ll either validate or discover a better idea hiding underneath.

The Full Story

The Overthinking Trap

John (a YC partner) nails the two ways founders paralyze themselves. First: hunting for the perfect idea in abstract. You can’t. Ideas only survive contact with reality and customer signal. Second: waiting for perfect founder-market fit. Blake Scholl built Boom Supersonic after adtech at Amazon—zero aviation background. Now it’s a billion-dollar company. Domain credentials matter less than curiosity plus depth.

Why Single-Threaded Beats Multithreading

Juggling three startup ideas produces noise, not signal. If you don’t go deep, you either ditch winners or double down on losers based on bad data. The fix: burn the boats. Kill the other ideas, rebrand, change emails, change how you talk about the mission. GovDash did this five times—each pivot felt like becoming a new person. By the fifth iteration (government procurement), they grew so fast they barely kept up and raised series B.

What “Going Deep” Actually Means

Can you run your customer’s business? Not “talked to 20 owners.” Can you step into a cleaning service tomorrow and know what their daily crises are? Which calls matter? What they’d pay to never lose another one? Could you teach a master class on the problem? The bar is: be one of the most informed people on Earth about the space.

Real learning happens in a tight loop: deep customer understanding → build → deeper understanding → better product. Real customers using your product give you concrete data that fills the gaps in pure abstract knowledge. Don’t wait for 100 interviews before writing code. Do both at once.

Three Qualities of Good Ideas (AI Era)

Edge of capability. Your product barely works on today’s frontier models but will clearly improve as they improve. Understand the bottlenecks intimately. If a bottleneck clears, solving it might be the company.

Verticalize. Sell an outcome (insurance, healthcare, banking) not software-for-X. Software cost goes to zero. Value lives in customer trust, licenses, regulatory moats, and owning the outcome. Corgi Insurance took this seriously: they acquired an actual insurance carrier during YC to own the full stack—underwriting to customer service. Now they underwrite any insurance line in any vertical with a fraction of traditional headcount.

Be the most ambitious version. The cost of a wildly ambitious idea and a modest one is roughly the same—both destroy your life. So swing for the version that, if it works, rewrites a sector. It also protects you from competitors, attracts talent, and builds a moat worth keeping.

What Failure Really Teaches

If you do all this and fail, you’re not back where you started. You have unambiguous customer data: is this actually a hair-on-fire problem or something you invented? You have real conviction for your next pivot and execution muscle. Most importantly: going deep reveals the better idea underneath. Founders start solving surface-level pain. The real opportunities are deeper structural problems. You’ll spot bottlenecks, gaps, missing dev tools. One of those becomes the actual company.

Key Takeaways

Overthinking kills startups. The perfect idea doesn’t exist in abstract. Pick one and collide with reality.
Single-minded focus produces signal. Juggling ideas produces noise and bad decisions. Commit fully.
Founder-market fit is not a prerequisite. Curiosity + depth + customer conversations beats credentials.
Can you run their business? The real test of deep knowledge: step into a customer’s shoes tomorrow and know their crisis points.
Verticalize, don’t build software-for-X. Own the outcome (insurance, healthcare, banking) not the software layer. Software is zero-cost; moats live elsewhere.
Ambitious ideas have the same cost as modest ones. If you’re going to suffer, swing for the version that rewrites a sector.
Failure reveals the better idea. Going deep isn’t validation. It’s reconnaissance. You’ll spot the deeper structural problem hiding under the surface one.
Signal per unit time compounds. Commit, walk fast, learn more per day than samplers learn per month.

Claude’s Take

This is John’s philosophy distilled to its useful core: commitment beats optionality in the fog. The logic is sound. When you don’t know which direction is up, walking fast in one direction teaches you more than tiptoeing in five.

The Corgi Insurance example is the standout. Most founders would “build software for insurance companies.” Corgi said “become the insurance company.” That mindset shift—verticalize, own the outcome, build the full stack—is the actually novel bit here, and the AI angle (fraction of traditional headcount) makes it concrete.

The other key insight: going deep isn’t primarily idea validation. It’s reconnaissance. You’ll discover the real problem (and thus the real company) hiding under the surface problem. This tracks with what works in practice: the best companies often started solving X, found Y was the real blocker, and pivoted to Y.

One mild note: John doesn’t address the case where your first pick was genuinely bad and you should have tried another option instead. The “burn the boats” framing leans all the way into commitment, which is usually right, but can create sunk-cost bias in some founders. The way out: commit hard for a season, then measure against reality with humility.

Score: 8. Practical, specific examples, clear mechanism (why depth beats breadth), and the verticalization insight carries weight.