{
  "video_id": "R56RJFZBasQ",
  "channel_slug": "ycombinator",
  "channel_handle": "ycombinator",
  "title": "Pick One Idea and Go Deep",
  "duration_seconds": 691.0,
  "url": "https://www.youtube.com/watch?v=R56RJFZBasQ",
  "upload_date": "",
  "transcript": "[music]\nHi, I'm John and I'm a partner at YC. I\noften meet founders who have lots of\nideas about what to work on and can't\ndecide between them. Sometimes they're\nworking on multiple things. Often\nthey'll say that they're waiting to find\nthe best idea before fully committing.\nBut it's extremely hard to make\nmeaningful progress on a startup without\ncommitting to a single idea. So in this\nvideo, I'm going to give you a rubric\nfor how to stop overthinking, pick an\nidea, commit to it, and then figure out\nfast whether it's actually working. The\nmost important piece of advice I'd give\nto founders struggling to pick a startup\nidea is don't overthink it. Overthinking\na startup in the earliest days can take\nmany forms, but here are a couple of the\nmost common failure modes I see. The\nfirst is thinking that you need to find\nthe perfect idea. In some ways, this is\nan understandable impulse. Startups are\nhard, so shouldn't you figure out the\nbest idea before committing? The problem\nwith this approach is that it's\nimpossible to figure out the perfect\nidea in the abstract. You can only\nfigure out what you should be working on\nby making contact with reality and\ngetting feedback from customers. The\nsecond overthink is, am I the perfect\nfounder for this? It's true that founder\nmarket fit matters. A non-technical\nfounder likely won't be the right person\nto come up with a killer DevTools\nstartup idea, for example. But often\nfounders, especially second-time\nfounders, weaponize this line against\nthemselves. They convince themselves\nthat they need a decade of domain\nexperience before they can start. The\ntruth is you don't. If you pick an idea\nyou're curious about, go extremely deep\nand most importantly talk to customers.\nIt's often possible to develop\nextraordinary knowledge in a short\namount of time. We see incredible\nexamples of this all the time at YC.\nTake Blake Scho, the CEO of Boom\nSupersonic. Blake spent his early career\nworking on adtech at companies like\nAmazon and Groupon before deciding to\nwork on commercializing supersonic\nflight. Lots of people probably thought\nhe was crazy. But now Boom is a\nbillion-dollar company. So don't let the\nquestion of whether you're allowed to\nwork on something stop you from\nstarting. Once you've stopped\noverthinking your ideas, it's important\nto commit to just one. Often I meet\nfounders who are working on multiple\nideas at once because they believe that\nthis is the best way to figure out which\none will actually work. There are a\ncouple problems with this approach. The\nmost serious is that it tends to produce\nbad data. If you don't actually go deep\non an idea, but instead juggle it with\nseveral others, you won't get good\nsignal about whether what you're doing\nactually works. And if you don't get\ngood signal, then you could either\nprematurely talk yourself out of a good\nidea or convince yourself that a bad one\nis worth continuing. The solution to\nthis is to go in depth first. If you're\ntrying to decide between several ideas,\nall of which look equally attractive,\npick one idea and go deep on it. What do\nI mean by going deep? The first thing is\nthat you should burn the other boats.\nThat is, you should explicitly foreclose\nyour other startup idea options, stop\nworking on them, tell any customers that\nyou've pivoted, and work with\nsingle-minded focus on the idea you've\nchosen. One way to think about going\ndeep is that it should feel like wearing\na new skin. You should become an almost\nunrecognizable version of yourself. This\ncould mean changing your company's name,\nyour emails, your website, and even your\ninternal narrative about why you're\nbuilding a startup in the first place.\nFor example, I worked with a startup\ncalled GovDash that helps customers win\ngovernment contracts. They pivoted at\nleast five times before finding this\nidea. And each time they explored\nsomething new, they changed their\ncompany name and how they talked about\ntheir mission. At one point, I forgot\nhow to get in touch with them because\nthey changed their email addresses with\neach pivot. By truly becoming domain\nexperts in government procurement, their\nfifth idea worked so well that they\ncould barely keep up with demand. They\nrecently raised the series B to scale\nthe business and meet that demand. Once\nyou've decided to fully commit to an\nidea and go deep, how do you know if\nyou're actually doing it well? The high\nwatermark I use to help founders answer\nthis question is, could you actually run\nyour customers business? Say you want to\nbuild voice customer service agents for\ncleaning services. The question isn't\njust whether you've talked to 20 owners.\nThe question is, if I dropped you into a\ncleaning business tomorrow, would you\nknow how to run it? Do you know what\ntheir daily crises are? Do you know\nwhether answering the phone is a top\nfive problem? Do you know how much\nbusiness they lose when a call goes\nunanswered and what they would actually\npay to never lose another one? These are\nthe kinds of questions you need to be\nable to answer with very high\nconfidence. Another way to think about\nthis is could you teach a class on the\nproblem you're solving? Are you one of\nthe most informed people in the world on\nthe subject? Getting to this level will\ninvolve lots of conversations with\ncustomers and sometimes even literally\ndoing the job yourself. But don't obsess\nover needing to talk to hundreds of\ncustomers before writing code. The goal\nis to do both at the same time in a\ntight loop. Deep understanding of\ncustomer needs, then product delivery,\nthen deeper understanding of customer\nneeds, then better product delivery.\nReal customers using your product\nproduces concrete data that will\ncomplement your abstract knowledge,\ngiving you a sense of whether what\nyou're building is actually working.\nOnce you're going deep on an idea,\nthere's several ways to validate whether\nit's worth continuing to work on. The\nmost obvious one is pull from customers,\nbut there are several other qualities of\ngood ideas in the AI era that you should\nlook out for as you go. The first is\nthat the idea sits at the edge of what\nmodels can do today. This might mean\nthat your product barely works on\ntoday's frontier models, but will\nclearly improve as they get better. You\nshould understand the bottlenecks\nimpeding your product's performance\nintimately if a particular bottleneck\ndoesn't clear the way you hoped. Solving\nthat might become the company. This is a\nversion of Paul Graham's well-known\nquote that you should live in the future\nand then build what's missing. The\nsecond quality of a good idea is that it\nshould verticalize. By this I mean that\nit should ultimately sell an outcome.\nfor example, providing insurance or\nmedical care rather than just software.\nIn the AI era, the cost of producing\nsoftware is going to zero. So the things\nthat actually become valuable aren't\njust software for X. They're customer\ntrust, licenses, regulatory permission,\nand outcome ownership. So if you want to\nget into the insurance space, don't\nbuild software for insurance companies.\nJust be the insurer. Similarly, rather\nthan selling back office software for\nbanks, just be the bank. One example of\nthis is Corgi Insurance, an AI powered\ncommercial insurance company from YC's\nsummer 24 batch. They were not content\nwith being a tech- enabled broker or\neven a managing general agent because\nthat was just owning a part of the\nsolution. Instead, they set an ambitious\ngoal of owning everything from\nunderwriting to providing customer\nservice, the entire commercial insurance\nstack, and even took the unprecedented\nstep of acquiring an insurance carrier\nduring their YC batch to make it happen.\nBeing the full stack insurance company\nallows Corgi to underwrite any insurance\nline in any vertical with a fraction of\nthe headcount of traditional carriers.\nThey can offer far better pricing, much\nfaster turnaround, and own all of the\neconomics. That brings me to the third\nquality of a good idea. It should be the\nmost ambitious version of itself. It may\nseem unintuitive, but the cost of\npursuing a wildly ambitious startup idea\nand the cost of pursuing a modest one\nare roughly the same. They're both\nextremely hard. They both place extreme\ndemands on your time. So aim at the\nversion that if it works rewrites a\nsector of the economy because that's\nalso the version that protects you from\ncompetitors, attracts the best talent\nand has a moat worth building. This\ncould mean building and selling into the\nmost regulated industries like legal,\nhealthcare, or financial services, or\ntaking on very large incumbents like a\n10 billion dollar legacy SAS company, or\nbuilding hard techch like robotics for\nspace assembly. Now, what if you do all\nof this and the idea fails? The good\nnews is that you'll be in a dramatically\nbetter position than where you started.\nFirst, you have unambiguous customer\ndata. You know whether there's actually\na hair on fire problem in this space or\nwhether you just talked yourself into\nthinking there was. You'll have real\nconviction to base a pivot on and a\nbetter sense of how to execute going\nforward. But more importantly, you will\noften come away from the process with a\nnew idea that will actually work. When\nmost founders begin, they're solving\nsurface level pain points. The real\nopportunities are almost always the\ndeeper structural problems. In other\nwords, going deep isn't primarily a\nprocess for validating the idea you\nstarted with. It's a way to find the\nbetter idea underneath. This almost\nalways happens, especially if you're at\nthe forefront of what models can do\ntoday. You'll notice the bottlenecks,\nthe gaps, the dev tools nobody's built,\nand one of those could turn out to be\nthe actual company. Here's what I want\nyou to take away from this video. First,\nstop trying to find the perfect idea.\nJust pick one. Then, burn the other\nboats. Learn everything you can about\nthe customer and try to execute for\nthem. In the early idea fog, where you\ncan only see 10 ft in front of you, the\ntemptation is to take a few cautious\nsteps in every direction. Sample a\nlittle here, a little there, stay close\nto home. The problem is that gives you\nalmost no information. What actually\nworks is to commit to one direction and\nwalk fast. You're not guaranteed to end\nup in the right place, but you generate\nmuch more information per unit of time.\nAnd when you're walking, you might\narrive at a better destination, one you\ncouldn't have seen from the start. The\nworst failure mode isn't being wrong.\nIt's not making a decision, spinning\nyour wheels, dabbling between ideas, and\nnever going deep enough on any one of\nthem to learn anything. So, pick one and\ngo deep. Thanks for watching. [music]\n[music]",
  "transcript_chars": 10691,
  "ingested_at": "2026-06-17T16:30:58.910378+00:00",
  "source": "channel",
  "yt_meta": {
    "view_count": null,
    "like_count": null,
    "channel_id": null,
    "categories": null,
    "tags": null
  }
}