Y Combinator - Pick One Idea and Go Deep

Short Summary

Main recommendation: pick one startup idea, commit to it fully, and go deep by talking to customers and learning the domain instead of waiting to find a perfect idea. The speaker argues that you can only learn whether an idea is right by making contact with reality, and that scattered effort produces bad data. He also argues that going deep often reveals a better, underlying problem worth building for.

Key Points

• Overthinking is a failure mode: waiting for the "perfect idea" is impossible because you can only find out what works through customer feedback.
• You do not need decades of domain experience to start. The speaker says it is often possible to build extraordinary knowledge quickly by going deep and talking to customers, citing Blake Scholl, CEO of Boom Supersonic, who moved from adtech to supersonic flight.
• Juggling several ideas produces bad signal, which can lead you to prematurely drop a good idea or keep a bad one.
• Strong ideas, in the speaker's view, sit at the edge of what AI models can do, verticalize into owning an outcome rather than just software, and aim at the most ambitious version of the problem.
• If an idea fails, going deep still leaves you with unambiguous customer data and often a better underlying problem to solve.

Details Worth Noting

• Concrete instructions:
  • Pick one idea, even if several look equally attractive.
  • Burn the other boats: explicitly drop your other startup options, stop working on them, and tell any customers you have pivoted.
  • Adopt a new identity for the chosen idea: change the company name, emails, website, and how you describe your mission.
  • Test whether you could run your customer's business: know their daily crises, whether answering the phone is a top five problem, how much business they lose on unanswered calls, and what they would pay to avoid losing it.
  • Test whether you could teach a class on the problem and whether you are among the most informed people on the subject.
  • Run customer conversations and product building in a tight loop, rather than talking to hundreds of customers before writing any code.
  • Check whether the product barely works on today's frontier models but will clearly improve as they get better, and understand the bottlenecks in detail.
  • Sell an outcome rather than software: for example, be the insurer rather than building software for insurers.
  • Aim for the version that, if it works, rewrites a sector, since the cost of pursuing an ambitious idea is roughly the same as a modest one.
• Examples: GovDash pivoted at least five times, changing its name and messaging each time, before its fifth idea—helping customers win government contracts—gained strong demand and raised a Series B. Corgi Insurance, from YC's summer 24 batch, acquired an insurance carrier to own underwriting and customer service end to end.
• Caveat: the speaker's own framing is that "could you run your customer's business" is a high bar and not a checkbox; the talk gives no numeric thresholds for how many customer conversations are enough.
• The speaker cites Paul Graham's idea to "live in the future and then build what's missing."

Bottom Line

Stop searching for the perfect startup idea. Choose one, commit fully, and walk fast in that direction, because the information you gain from deep focus is worth more than cautious exploration of many options.