YC Partner Jon Xu Urges Founders to Pick One Startup Idea and Burn the Boats

Y Combinator

Deep commitment to a single initial concept accelerates customer discovery and exposes the stronger underlying business opportunities hidden beneath surface-level problems.

AI-assisted summary of Pick One Idea and Go Deep — original source. Check the original for full context.

Source published: 2026-06-17

Detailed notes

## Background This video was published on YouTube by Y Combinator on June 17, 2026. In this episode of Startup School, YC General Partner Jon Xu explains why early-stage founders should avoid getting paralyzed looking for the perfect startup idea, advising them instead to pick one concept, commit fully, and go deep with real customers.

## TL;DR Jon Xu addresses the common 'idea fog' that paralyzes founders trying to pick the best startup concept. He argues that trying multiple ideas at once yields low-quality feedback, whereas picking one and 'burning the other boats' allows founders to gain true domain authority and discover deeper underlying problems. To succeed, founders must understand customer operations deeply enough to run their businesses, build at the edge of current AI capabilities, verticalize to own full outcomes, and aim for highly ambitious goals. Ultimately, moving fast in a single direction produces actionable information, while indecision remains a startup's worst failure mode.

## Atmosphere & style The presentation is direct, highly pragmatic, and authoritative, delivering clear strategic guidance with a grounded Y Combinator perspective. Xu uses vivid analogies—like 'wearing a new skin' or walking fast through an 'idea fog'—to make abstract founder dilemmas feel actionable and concrete. The overall mood is motivating yet uncompromising, pushing founders to abandon cautious overthinking in favor of relentless execution and deep customer focus.

## Key takeaways - [00:00](https://www.youtube.com/watch?v=R56RJFZBasQ&t=0s) Founders often fall into the 'perfect idea trap' or weaponize founder-market fit against themselves, delaying progress instead of making contact with reality and talking to customers. - [01:22](https://www.youtube.com/watch?v=R56RJFZBasQ&t=82s) Domain expertise can be built rapidly simply by following curiosity, going deep, and constantly seeking customer feedback. - [02:42](https://www.youtube.com/watch?v=R56RJFZBasQ&t=162s) Working on multiple ideas simultaneously yields noisy, low-quality data; founders must commit to 'in-depth first' and 'burn the other boats.' - [03:21](https://www.youtube.com/watch?v=R56RJFZBasQ&t=201s) Going deep requires completely stepping into a new skin, updating company branding, email domains, and internal narrative to build genuine domain authority. - [04:51](https://www.youtube.com/watch?v=R56RJFZBasQ&t=291s) A key test for customer understanding is whether a founder is knowledgeable enough about daily operational crises to run the customer's business or teach a college course on it. - [06:21](https://www.youtube.com/watch?v=R56RJFZBasQ&t=381s) Product development and customer dialogue must run in a tight, iterative execution loop rather than holding off on code until hundreds of interviews are completed. - [06:18](https://www.youtube.com/watch?v=R56RJFZBasQ&t=378s) In the modern AI landscape, startups should build at the absolute edge of model capabilities so that product value scales alongside frontier model improvements. - [07:15](https://www.youtube.com/watch?v=R56RJFZBasQ&t=435s) Because AI is driving software production costs to zero, defensible value now lies in verticalizing to sell complete outcomes, owning licenses, regulatory status, and customer trust. - [08:37](https://www.youtube.com/watch?v=R56RJFZBasQ&t=517s) Founders should aim for the most ambitious version of their idea, as targeting massive economic sectors costs similar energy while attracting better talent and building stronger moats. - [09:33](https://www.youtube.com/watch?v=R56RJFZBasQ&t=573s) Going deep is the best mechanism to uncover structural realities, allowing founders to pivot from surface-level ideas to the far superior idea hidden underneath. - [10:27](https://www.youtube.com/watch?v=R56RJFZBasQ&t=627s) The primary failure mode in early-stage startups is indecision; picking one direction and walking fast produces maximum actionable information.

## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=R56RJFZBasQ&t=0s) Intro & The "Perfect Idea" Trap

Y Combinator General Partner Jon Xu works with many startup founders who are stuck in "idea fog"—paralyzed by having multiple startup concepts and struggling to choose between them. Founders often hold off on fully committing because they are waiting to find the single "best" idea first. However, making meaningful startup progress is nearly impossible without single-minded commitment. Overthinking in the early days usually takes two common forms: searching for the perfect idea in the abstract, and questioning whether one has the ideal background to build it.

Trying to evaluate an idea purely in theory fails because real insight only comes from making contact with reality and getting feedback from actual customers. Similarly, founders—especially second-time founders—often "weaponize" founder-market fit against themselves, assuming they need a decade of domain expertise before launching. While technical fit matters for specialized tools, founders can develop extraordinary domain knowledge in a short period simply by following their curiosity, going deep, and talking to customers.

- **Example — Boom Supersonic:** CEO Blake Scholl spent his early career in ad tech at Amazon and Groupon before setting out to commercialize supersonic passenger flight. Though many thought he was crazy, Boom became a billion-dollar company.

> "You can only figure out what you should be working on by making contact with reality and getting feedback from customers."

---

[02:42](https://www.youtube.com/watch?v=R56RJFZBasQ&t=162s) Why Working on Multiple Ideas Fails & How to Go Deep

Dividing focus across multiple ideas to see which one works generates noisy, low-quality data. Without going deep, founders miss genuine signal, leading them to abandon great concepts prematurely or double down on flawed ones. The solution is "in-depth first": pick one attractive option, commit completely, and "burn the other boats" by explicitly foreclosing all other paths.

Going deep means stepping into a whole new skin—becoming an almost unrecognizable version of yourself by changing company names, email addresses, landing pages, and the internal narrative of why the startup exists. Founders must operate with single-minded focus to build real domain authority.

- **Example — GovDash:** A startup helping companies win government contracts pivoted at least five times before hitting on its winning concept. With each pivot, they updated their company name, email domain, and mission statement—to the point where Xu temporarily lost track of how to contact them. By immersing themselves in government procurement, their fifth pivot took off, leading to explosive demand and a Series B raise.

> "Going deep... should feel like wearing a new skin. You should become an almost unrecognizable version of yourself."

---

[04:51](https://www.youtube.com/watch?v=R56RJFZBasQ&t=291s) Could You Run Your Customer's Business?

To evaluate depth of commitment, Xu uses a practical litmus test: *Could you actually run your customer's business tomorrow?* If building voice AI agents for cleaning services, talking to 20 business owners is not enough. You must understand their daily operational crises, whether missed calls are a top-five issue, how much revenue vanishes when a call drops, and what they would pay to fix it. Alternatively, ask if you could teach a college class on the exact problem or if you are among the most knowledgeable people on the topic worldwide.

Achieving this depth requires constant customer dialogue and sometimes doing the job manually. However, founders should not wait to write code until they have interviewed hundreds of people. Product development and customer feedback must run concurrently in a tight, iterative feedback loop: customer understanding leads to product delivery, which yields deeper customer insights and a better product.

- **Key Question:** Are you knowledgeable enough about your customer's daily operational pain points to teach a class on the subject? - **Execution Loop:** Deep Customer Understanding $\rightarrow$ Product Delivery $\rightarrow$ Deeper Insight $\rightarrow$ Refined Product Delivery.

---

[06:18](https://www.youtube.com/watch?v=R56RJFZBasQ&t=378s) Validating Ideas: Building at the Edge of AI & Verticalizing

When validating whether an idea is worth pursuing, founders should look beyond basic customer pull for specific qualitative indicators that are crucial in the modern AI landscape. First, build at the absolute edge of model capabilities—where a product barely functions on today's frontier models but will naturally scale as underlying AI improves. If a technical bottleneck impedes progress, solving that bottleneck might become the core business (echoing Paul Graham’s famous mantra to "live in the future and build what's missing").

Second, the idea should verticalize by selling complete outcomes (e.g., providing actual insurance policies or medical care) rather than basic software. Because AI is driving the marginal cost of software production toward zero, raw code is no longer the value capture point. Value now lies in owning regulatory approvals, licenses, customer trust, and full-stack outcome delivery.

- **Example — Corgi Insurance (YC S24):** Rather than remaining a tech-enabled broker or managing general agent (MGA), Corgi set out to own the entire commercial insurance stack from underwriting to customer service. They even acquired an insurance carrier during their YC batch. This full-stack approach allows Corgi to underwrite any insurance line with a fraction of traditional headcount, offering better pricing, faster turnarounds, and full ownership of unit economics.

> "In the AI era, the cost of producing software is going to zero... The things that actually become valuable aren't just software for X. They're customer trust, licenses, regulatory permission, and outcome ownership."

---

[08:37](https://www.youtube.com/watch?v=R56RJFZBasQ&t=517s) Aim at the Most Ambitious Version & Unlocking the Idea Underneath

Pursuing a modest startup idea costs roughly the same amount of time, energy, and stress as pursuing a wildly ambitious one. Therefore, founders should target the version that rewrites an entire economic sector if it succeeds. Ambitious goals attract top talent, defend against incumbents, and build enduring moats—whether that means navigating heavily regulated sectors (legal, healthcare, finance), taking on $10B SaaS incumbents, or building hardtech like space assembly robotics.

Even if the initial idea fails, going deep puts founders in a superior position. It yields concrete customer data, clear signal on whether a problem is truly "hair-on-fire," and real conviction for a pivot. Most importantly, initial ideas usually address surface-level symptoms; digging deep exposes the underlying structural problems where true startup opportunities lie.

- **Key Insight:** Going deep is rarely just about validating the original idea—it is the fastest mechanism to discover the far better idea hidden beneath it. - **AI Ecosystem Opportunity:** Operating at the edge of AI capabilities inevitably exposes missing infrastructure, operational bottlenecks, and developer tools that need to be built.

---

[10:27](https://www.youtube.com/watch?v=R56RJFZBasQ&t=627s) Walk Fast in One Direction

In the early "idea fog," where visibility is limited to ten feet, the natural instinct is to take cautious steps in every direction. However, sampling options without committing generates almost zero actionable information. The winning playbook is picking one direction and walking fast. Moving briskly produces maximal information per unit of time and brings founders to valuable destinations they could not have anticipated from the start.

- **The Primary Failure Mode:** The worst outcome in a startup is not being wrong; it is failing to make a decision—spinning wheels, dabbling across ideas, and never going deep enough anywhere to learn anything. - **Core Directive:** Stop searching for the perfect idea. Pick one, burn the other boats, and go deep.

## Notable quotes > [00:00](https://www.youtube.com/watch?v=R56RJFZBasQ&t=0s) You can only figure out what you should be working on by making contact with reality and getting feedback from customers. — Jon Xu

> [02:42](https://www.youtube.com/watch?v=R56RJFZBasQ&t=162s) Going deep... should feel like wearing a new skin. You should become an almost unrecognizable version of yourself. — Jon Xu

> [04:51](https://www.youtube.com/watch?v=R56RJFZBasQ&t=291s) Could you actually run your customer's business tomorrow? — Jon Xu

> [06:18](https://www.youtube.com/watch?v=R56RJFZBasQ&t=378s) In the AI era, the cost of producing software is going to zero... The things that actually become valuable aren't just software for X. They're customer trust, licenses, regulatory permission, and outcome ownership. — Jon Xu

## Mentioned Y Combinator, Jon Xu, Boom Supersonic, Blake Scholl, Amazon, Groupon, GovDash, Corgi Insurance, Paul Graham

[Source](https://www.youtube.com/watch?v=R56RJFZBasQ)

中文简报

YC合伙人Jon Xu敦促创始人选定单一创业想法并破釜沉舟对单一初期构想的深度投入能加速客户发现，并揭示隐藏在表面问题之下的更强大潜在商业机会。 Jon Xu探讨了常常使试图挑选最佳创业概念的创始人陷入瘫痪的“想法迷雾”。他认为，同时尝试多个想法只会获得低质量的反馈，而选定一个并“破釜沉舟”能让创始人获得真正的行业权威，并发现更深层的根本问题。若想取得成功，创始人对客户业务运营的理解必须足够深刻，甚至达到能够代为经营客户企业的程度；必须在当前AI能力的前沿进行构建，通过垂直化掌控完整成果，并树立极具野心的目标。归根结底，朝单一方向快速前进能产生切实可行的信息，而优柔寡断仍是初创公司最糟糕的失败模式。创始人往往会陷入“完美想法陷阱”，或者将“创始人与市场契合度”当作对抗自己的武器，不仅未能接触现实并与客户交流，反而拖延了进展。 (0:00) 仅凭追随好奇心、深入探索并持续寻求客户反馈，便能迅速建立起行业专长。 (1:22) 同时推进多个想法会带来充满噪音的低质量数据；创始人必须践行“深度优先”并“破釜沉舟”。 (2:42) 深入探索需要彻底换上新的身份，更新公司品牌、邮箱域名以及内部叙事，以此建立起真正的行业权威。 (3:21) 衡量对客户理解程度的一个关键检验标准，是创始人对日常运营危机的熟悉程度是否足以让他们经营客户的公司，或就该领域讲授一门大学课程。 (4:51)

Reading summary

Y Combinator General Partner Jon Xu argues that early-stage founders fail by dabbling across multiple ideas instead of committing fully to one. Meaningful progress requires single-minded focus, rapid customer contact, and diving deep enough to uncover the structural problems hidden beneath surface concepts. The Trap of Abstract Evaluation and Multitasking (0:00) Founders frequently stall in an 'idea fog,' trying to select the perfect concept in theory or questioning whether they possess sufficient domain background. Xu notes that founders often weaponize founder-market fit against themselves, mistakenly believing they need a decade of specialized experience before launching. In practice, deep curiosity and direct customer conversations rapidly bridge knowledge gaps. For example, Boom Supersonic CEO Blake Scholl moved from an ad tech career at Amazon and Groupon to build a billion-dollar supersonic aircraft venture. Committing Completely to Filter Noisy Signals (2:42) Dividing effort among several simultaneous ideas produces low-quality, noisy feedback that leads founders to abandon viable ideas prematurely or persist with poor ones. Xu advises an 'in-depth first' approach: selecting one option, burning alternative paths, and completely immersing oneself in that problem space. For instance, GovDash cycled through five distinct pivots while pursuing government contracting software—updating its company name, domain, and mission each time—until fully immersing in procurement processes unlocked explosive demand and a Series B round. The Operational Litmus Test for Customer Understanding (4:51) To measure true problem immersion, Xu asks founders whether they could step in and run their customer's business the next day, or teach a college course on the topic. Simply surveying a few dozen prospects is insufficient; founders must grasp granular operational realities, such as the exact revenue lost when a service business misses an inbound call. However, deep customer research should not delay technical execution. Building and customer dialogue must run as a tight, concurrent loop where delivering initial product iterations unlocks deeper operational insights. Capturing Value at the Frontier of AI (6:18) When vetting startup concepts in the current technology environment, Xu highlights two qualitative criteria: operating at the frontier of AI capabilities and pursuing full-stack verticalization. Founders should build where software barely functions on existing models but will naturally benefit from model improvements. Furthermore, because AI drives the marginal cost of producing raw software toward zero, lasting value shifts from software interfaces to owning complete outcomes, regulatory licenses, and customer trust. Illustrating this shift, YC company Corgi Insurance acquired a carrier to control its entire commercial insurance stack rather than remaining a tech-enabled broker. Maximizing Ambition to Reveal Root Problems (8:37) Because modest startup concepts demand the same time and emotional toll as transformative ones, Xu recommends pursuing the most ambitious version of an idea—such as entering regulated sectors or building hardtech. Going deep rarely just validates the starting premise; digging beneath surface symptoms is the fastest way to expose the underlying structural bottleneck worth solving. Xu contends that moving fast in one direction yields maximal information per unit of time, whereas dabbling cautiously yields almost no actionable signal. According to Xu, the primary failure mode in early-stage startups is not pursuing the wrong concept, but succumbing to indecision. Even if an initial hypothesis fails, committing completely and moving aggressively provides the operational clarity and customer evidence required to pivot effectively.

精炼阅读摘要

Y Combinator 合伙人 Jon Xu 指出，初创团队最易陷入在多个构想间摇摆不决的“想法迷雾”陷阱；创业者应切断其他退路并全身心深潜于单一构想，通过与现实客户的高频碰撞获取真实信号，进而挖掘深层商业机会。摒弃理论评估与背景执念 (0:00) 许多创业者推迟做决策，试图在抽象理论层面推导最完美的构想，或过度怀疑自身是否具备足够的行业履历。Jon Xu 认为，仅凭纯理论推演无法获得有效的商业洞察，认知必须建立在与现实世界接触以及真实客户反馈的基础之上。连环创业者尤其容易将“创始人与市场契合度”（Founder-Market Fit）当成自我设限的借口，误以为必须先具备十年的行业经验才能启动项目。除了高度专业化的底层技术工具外，创业者完全可以在好奇心驱使下，通过深入钻研与大量客户对话，在短时间内构建起深厚的行业认知。例如，Boom Supersonic 创始人 Blake Scholl 早期职业生涯主要在亚马逊和 Groupon 从事广告技术，但他跨界投身超音速民航客机的商业化开发，最终将公司打造为估值超十亿美元的独角兽企业。破釜沉舟与彻底重塑身份 (2:42) 在多个构想之间分散精力以测试可行性，往往只能收集到充斥噪音的低质量数据。由于缺乏足够的投入深度，创业者容易错失真正的市场信号，导致其过早放弃优秀构想，或在存在致命缺陷的方向上盲目追加投入。Jon Xu 主张采取“深度优先”策略：选定一个具吸引力的选项并全力以赴，通过彻底切断其他潜在路径来“破釜沉舟”。深潜意味着创业者需要彻底重塑自我形态，像换上一层“新皮肤”一样全面调整公司名称、邮箱域名、落地页与创业叙事，从而以专一的姿态建立起行业话语权。例如，专注于协助企业竞标政府采购合同的 GovDash，在最终确立商业模式前经历了至少五次转型。该团队在每次调整时都彻底重置公司名称、邮箱和使命宣言，甚至一度令合伙人失去联系方式；正是依靠对政府采购流程的彻底沉浸，其第五次转型迎来了爆发式增长并顺利完成 B 轮融资。运营级认知与敏捷迭代闭环 (4:51) 为了衡量创业者对业务的理解深度，Jon Xu 提出了一个检验标准：你明天能否直接接管并运营客户的企业？如果团队正在为清洁服务行业开发语音 AI 代理，仅仅与 20 位经营者交谈远远不够。创始人必须弄清客户每日面临的实际运营危机、漏接电话是否属于其前五大痛点、每漏接一次会损失多少收入，以及对方愿意支付多少资金来解决这一问题；或者反思自己是否足够熟悉该领域，能够开设专门的大学课程。与此同时，深潜并不意味着要完成数百次访谈后才开始编写代码。产品开发与客户调研必须紧密交织并同步推进，形成敏捷的迭代闭环：深度的客户认知驱动精准的产品交付，而产品交付又会激发出更深层的客户反馈，进而促成产品的持续优化。站在模型边界交付完整结果 (6:18) 在现代 AI 格局下评估构想的价值时，创业者除了关注基础的客户需求外，还需遵循两项关键原则。首先，产品应当直接构建在当前模型能力的最前沿边缘——即便在当前的前沿模型上仅能勉强运行，但能随着底层 AI 能力的迭代自然扩展；如果研发遭遇技术瓶颈，攻克该瓶颈本身便可能成为核心业务护城河。其次，商业构想必须向垂直化演进，出售最终的业务结果（例如直接提供保险单或医疗服务），而非单纯提供软件工具。Jon Xu 指出，AI 正在将软件代码的边际生产成本压低至趋近于零，因此单纯的代码已不再具备价值捕获能力，核心壁垒已转移至客户信任、行业牌照、监管审批以及端到端的结果所有权。例如 YC S24 批次的 Corgi Insurance，其并未止步于科技型中介平台，而是在孵化期间直接收购了一家持牌保险承运机构，实现商业保险承保全流程的自营化，以极少的人力成本换取了更快的周转速度、定价优势以及对单位经济模型的完整掌控。追求宏大目标以击穿底层痛点 (8:37) 追求平庸的构想与追求宏大的构想所消耗的时间、精力和精神成本大致相当。因此，创业者应该直接瞄准那些一旦实现便能重塑整个行业经济格局的宏大方案。雄心勃勃的愿景有助于吸引顶尖人才，并在强监管行业（如法律、医疗、金融）、硬科技领域（如太空装配机器人）或直接对抗百亿美元 SaaS 巨头时构筑持久壁垒。即使初始构想最终失败，全力深潜也能让团队处于更有利的地位：它能提供真实的客户数据，验证痛点是否足够紧急，并为后续转型提供明确依据。创始人最初设想的方案通常只是浮于表面的症状，只有持续向深处钻研，才能击穿表象，暴露出隐藏在底层、真正蕴藏巨大创业契机的结构性问题。在早期能见度极低的“想法迷雾”中，多方尝试往往无法产生有效的决策信息；快速沿着单一方向推进，才能在单位时间内换取最大量的信息反馈。对初创团队而言，首要的失败模式并不是被证明犯错，而是在犹豫不决中轮番浅尝，从而从未深入任何领域以获取真正的突破。

Detailed notes — zh

## Background 本文来自 Y Combinator（YC）于 2026 年 6 月 17 日在 YouTube 平台上发布的视频《Pick One Idea and Go Deep》。视频由 YC 通用合伙人 Jon Xu 主讲，旨在为创业者解答如何打破寻找“完美想法”的思维僵局、如何“破釜沉舟”集中资源深耕单一方向，以及如何通过深入理解客户需求来挖掘更有价值的商业机会。 ## TL;DR 视频深入剖析了初创企业创业者常见的“完美想法”与“完美创始人”误区，指出同时推进多个项目会导致信号失真与决策失误。Jon Xu 强调，创业者应当挑选一个有潜力的方向并破釜沉舟式地全力投入，达到能够代管客户业务的认知深度。在 AI 时代，优质的创业想法应具备紧贴大模型前沿能力、全栈垂直化交付结果以及追求极高雄心壮志等特征；即便初始项目失败，深度扎根所获取的真实认知与结构性洞察，也能指引创业者发现深层次的巨大商机。 ## Atmosphere & style 整段演讲风格干练、务实且充满干货，展现出 YC 一贯的敏锐洞察与结果导向色彩。讲者以接地气的案例和针针见血的批判语气，精准戳中创业者在起步阶段容易出现的摇摆与焦虑。整体氛围既有警醒迷茫者的直白硬核，又洋溢着鼓励创业者勇于探索前沿、追求宏大野心的激情与力量。 ## Key takeaways - [00:59](https://www.youtube.com/watch?v=R56RJFZBasQ&t=59s) 试图在抽象层面凭空构建“完美想法”是常见的创业陷阱，真正的可行方向唯有通过与真实客户互动和获取反馈才能验证。 - [02:00](https://www.youtube.com/watch?v=R56RJFZBasQ&t=120s) 不要被“创始人与市场契合度”限制，行业领域知识可以通过选择充满好奇心的领域并在极短时间内深度深耕来快速弥补。 - [02:42](https://www.youtube.com/watch?v=R56RJFZBasQ&t=162s) 同时推进多个项目会导致精力和资源分散，产生劣质的信号与数据，进而引发严重的决策失误。 - [03:21](https://www.youtube.com/watch?v=R56RJFZBasQ&t=201s) 面对多个备选项目时应当“破釜沉舟”，选择一个全力投入，并彻底重塑公司的品牌、名称与使命。 - [04:51](https://www.youtube.com/watch?v=R56RJFZBasQ&t=291s) 衡量认知深度的标准是能否深刻理解客户痛点并替其运营公司，甚至达到能够针对该痛点开课教学的水平。 - [05:30](https://www.youtube.com/watch?v=R56RJFZBasQ&t=330s) 避免过度调研引发的教条主义，应建立“深度理解需求 $\rightarrow$ 交付产品 $ ightarrow$ 进一步理解需求 $ ightarrow$ 迭代”的快速闭环。 - [06:18](https://www.youtube.com/watch?v=R56RJFZBasQ&t=378s) AI 时代的项目应构建在模型能力的前沿边缘，使其能随着底层模型的指数级升级而自然进化。 - [07:27](https://www.youtube.com/watch?v=R56RJFZBasQ&t=447s) AI 时代软件生产成本趋近于零，创业项目应向全栈垂直化扩展，通过掌控信任、牌照与合规来直接出售最终结果。 - [08:37](https://www.youtube.com/watch?v=R56RJFZBasQ&t=517s) 追求平庸与宏大目标的艰辛程度相当，应选择最雄心勃勃的项目版本，以重塑行业并建立坚实的护城河。 - [09:33](https://www.youtube.com/watch?v=R56RJFZBasQ&t=573s) 即使项目最终失败，深耕也能带来无歧义的硬数据，并帮助创业者挖掘出隐藏在底层、真正可行的商业机会。 - [10:27](https://www.youtube.com/watch?v=R56RJFZBasQ&t=627s) 创业初期面对迷雾应选定单一方向快速迈进，以最大化产出有效信息；最糟糕的失败模式是犹豫不决、从不深度扎根。 ## Detailed walkthrough ### [00:00](https://www.youtube.com/watch?v=R56RJFZBasQ&t=0s) 告别“完美想法”与“完美创始人”的思维陷阱初创企业的创业者在刚起步时，最容易陷入的误区就是反复犹豫、无法决定该从事哪个项目。许多人甚至同时推进多个项目，声称必须找到“最佳想法”后才肯全力投入。但初期的过度思考往往会导致行动瘫痪。第一个常见的失败模式，是执念于寻找“完美想法”。创业固然充满挑战，但完美想法绝对无法通过抽象的空想得来；创业者必须与现实世界真实接触，从客户那里获取反馈，才能明确方向。第二个误区则是陷入“我是否是该项目的完美创始人”这种内耗。创始人与市场的契合度（Founder-market fit）固然重要，例如非技术背景的创始人确实很难打造出杀手级的开发者工具，但很多创业者（尤其是二次创业者）容易把这一点当成阻碍自己的武器，误以为必须拥有十年的行业经验才能开局。事实上，只要挑选一个充满好奇心的领域并极速深入，短时间内就能积累出惊人的专业知识。以 Boom Supersonic 的 CEO Blake Scholl 为例，他在创业初期仅在 Amazon 和 Groupon 等公司从事广告技术（ad tech）工作，后来却决定投身商业超音速飞行领域。当时很多人觉得他疯了，但如今 Boom 已经成长为一家价值数十亿美元的巨头公司。 - 无法在抽象层面凭空构想出完美想法，唯有通过真实的客户反馈才能验证。 - 领域经验可以通过深耕快速弥补，无需被“创始人与市场契合度”绑架。 - 案例：Boom Supersonic CEO Blake Scholl 从广告技术领域转战超音速客机，成功打造出十亿美元级别的公司。 ### [02:42](https://www.youtube.com/watch?v=R56RJFZBasQ&t=162s) 破釜沉舟：为什么必须“一次只押注一个想法” 不少创业者喜欢同时推进多个项目，以为这样能更快找出哪个方向可行。这种多管齐下的做法会产生极其严重的负面效应：由于精力和资源被分散，你无法在任何一个项目上深度扎根，从而导致获取的信号和数据极不准确。劣质数据会带来严重的决策失误，要么让你过早放弃一个潜力巨大的好想法，要么让你误以为某个糟糕的想法值得继续坚持。正确的策略是“先深入”——当面对几个看起来同样吸引人的选项时，果断挑选其中一个并全力深耕。所谓“深入”，意味着要“破釜沉舟”（burn the other boats），明确放弃其他备选方案。停止其他项目的研发，通知既有客户公司已转型，将全部精力聚焦于唯一的方向。这种深耕的过程应当像“换上一层全新的皮肤”，甚至让你变成一个全新的自己。这不仅包括变更公司名称、官方网站和电子邮箱，更意味着重新定义创业的底层使命。以政府合同竞标服务商 GovDash 为例，他们在找到最终方向前经历了至少 5 次转型。每次探索新方向，团队都会全面更换公司名称和使命表述，甚至连邮箱也跟着换，导致 YC 合伙人一度找不到他们的联系方式。但正是因为彻底成为了政府采购领域的专家，他们的第 5 个想法爆发出了极强的市场需求，近期已成功完成 B 轮融资。 - simultaneous 推进多想法会导致信号失真，极易误判项目的真实价值。 - “深入”需要全面重塑公司形态，包括公司名、官网、联系邮箱及内部叙事。 - 案例：GovDash 在历经至少 5 次彻底转型并随之更改品牌后，深度掌握政府采购门道，爆火并完成 B 轮融资。 ### [04:41](https://www.youtube.com/watch?v=R56RJFZBasQ&t=281s) 试金石测试：你能够代管客户的公司吗？衡量自己是否深耕到位的高标准试金石是：“如果明天把你扔进客户的公司，你能够替他运营这家公司吗？”假设你的创业方向是为清洁服务公司提供语音 AI 客服，关键不在于你是否访谈了 20 位老板，而在于你是否真正掌握了他们日常面临的危机、接听电话是否属于前五大痛点、漏接电话会造成多少损失，以及他们愿意付多少钱来一劳永逸地解决问题。创业者需要达到“能够针对该痛点开课教学”的认知深度，甚至在必要时亲自定义并亲自替客户打工体验。但与此同时，切忌陷入“写代码前必须访谈数百个客户”的教条主义。正确的节奏是构建一个紧密的闭环：深度理解客户需求 $\rightarrow$ 交付产品 $\rightarrow$ 更深入理解需求 $\rightarrow$ 迭代产品。真实用户的实际使用能提供具体的硬数据，与抽象认知相互补充，帮助你评估产品是否有效。 - 衡量认知深度的标准：能够代为运营客户的公司，或针对该业务痛点开课教学。 - 拒绝纸上谈兵，将深度客户调研与产品交付紧密结合，形成快速闭环迭代。 ### [06:18](https://www.youtube.com/watch?v=R56RJFZBasQ&t=378s) AI 时代好想法的三大特征在深度探索项目的过程中，评估一个想法是否值得长期坚持，除了看客户拉力（customer pull）之外，在 AI 时代还需要重点考察三大特质。首先，想法必须建立在“当前模型能力的前沿边缘”（edge of what models can do today）。这意味着产品在现有的前沿模型下可能仅能勉强运行，但随着底层模型的升级，产品效果会呈指数级提升。创业者必须极其敏锐地掌握限制产品性能的瓶颈；如果某个技术瓶颈未能如预期般被解决，攻克该瓶颈本身或许就会演变成公司的主营业务（此即 Paul Graham 所说的“生活在未来，然后打造缺失的部分”）。其次，好的想法应当具备“垂直化扩展”（verticalize）的能力，即直接出售最终结果（outcome），而非仅仅卖软件。在 AI 时代，软件的生产成本正在趋近于零，单纯的工具软件价值大幅缩水，真正的核心价值转向了客户信任、牌照、合规许可以及对结果的掌控。与其为保险公司开发软件，不如直接成为保险公司；与其为银行开发后勤软件，不如直接成为银行。以 YC 2024 年夏季营（S24）的项目 Corgi Insurance 为例，这家 AI 商业保险公司不满足于仅做技术赋能的经纪人或 MGA（代理商），而是树立了掌控从核保到客服全流程商业保险栈的宏大目标，甚至在 YC 训练营期间破天荒地收购了一家持牌保险公司（insurance carrier）。全栈化的保险公司形态让 Corgi 能够以传统承保商极小比例的人力成本，承保任何垂直领域的保险产品，提供更优质的定价与极速的响应。最后，想法应当选择“最雄心勃勃的版本”。追求一个野心勃勃的想法与追求一个平庸的想法，所付出的艰辛和时间成本几乎是一样的。创业者应当瞄准那个“一旦成功就能重塑某个经济领域”的宏大目标。只有极其雄勃的目标才能帮你抵御竞争对手、吸引顶尖人才，并建立起真正有价值的护城河。这包括深入法律、医疗、金融等高监管行业，挑战数百亿美元规模的传统 SaaS 巨头，或是开发太空组装机器人等硬科技。 - **模型前沿边缘**：建立在当前模型能力的极界，随着大模型升级而自动进化（践行 Paul Graham 的理念）。 - ** outcomes 交付与全栈垂直化**：直接出售结果并掌控全流程（信任、牌照、合规）。 - **案例（Corgi Insurance）**：YC S24 批次项目，收购保险承保商打通全栈商业保险，以极低人力实现极速核保与优质定价。 - **追求最高雄心版本**：平庸与宏大的创业难度相当，选择能重塑行业、吸引顶级人才并建立护城河的极高目标（如高监管行业、万级 SaaS 替代、硬科技等）。 ### [09:13](https://www.youtube.com/watch?v=R56RJFZBasQ&t=553s) 快速朝一个方向迈进：失败背后的真正价值如果全力投入后项目依然失败了，创业者依然处于比起点好得多的位置。首先，你获得了毫无歧义的客户数据，能够极其清醒地判断该领域是否存在“烫手山芋”般的刚需，为后续的转型（pivot）提供坚实的依据。更重要的是，深入探索的过程往往会孕育出真正可行的全新想法。绝大多数创始人刚起步时关注的只是表层的痛点，而真正的商业机会几乎都隐藏在更深层的结构性难题中。深耕的意义并不在于验证最初的那个雏形，而在于“挖掘出藏在底层更好的想法”。当你置身于技术前沿时，你会一眼洞察到那些尚未被满足的瓶颈、缺失的工具链，而这些工具本身可能就会演变为下一家巨头公司。在创业初期的迷雾中，能见度或许只有十英尺，试图向各个方向谨慎试探只会白白浪费时间且毫无收获。唯有选定一个方向“快速迈进”（walk fast in one direction），才能在单位时间内产出最多的有效信息。即便最终没有到达预设的终点，也会意外收获沿途更好的风景。创业中最糟糕的失败模式不是犯错，而是“不作决定”——在多个想法之间摇摆不定，却从未在任何一个方向上深度扎根。 - 深度深耕即使失败，也能获得明确的底层结构性认知，并探寻到隐藏在下方真正可行的商业机会。 - 创业初期如身处迷雾，向多方向试探无法获取有效信息，唯有单向快速推进才能最大化信息产出。 - 避开最恶劣的失败模式：因犹豫不决而在多个选项间打转、无法深耕。 ## Notable quotes > [00:00](https://www.youtube.com/watch?v=R56RJFZBasQ&t=0s) 第一个常见的失败模式，是执念于寻找“完美想法”。创业固然充满挑战，但完美想法绝对无法通过抽象的空想得来；创业者必须与现实世界真实接触，从客户那里获取反馈，才能明确方向。 — Jon Xu > [00:00](https://www.youtube.com/watch?v=R56RJFZBasQ&t=0s) 挑选一个充满好奇心的领域并极速深入，短时间内就能积累出惊人的专业知识。 — Jon Xu > [02:42](https://www.youtube.com/watch?v=R56RJFZBasQ&t=162s) 由于精力和资源被分散，你无法在任何一个项目上深度扎根，从而导致获取的信号和数据极不准确。 — Jon Xu > [02:42](https://www.youtube.com/watch?v=R56RJFZBasQ&t=162s) 当面对几个看起来同样吸引人的选项时，果断挑选其中一个并全力深耕。所谓“深入”，意味着要“破釜沉舟”（burn the other boats），明确放弃其他备选方案。 — Jon Xu > [04:41](https://www.youtube.com/watch?v=R56RJFZBasQ&t=281s) 如果明天把你扔进客户的公司，你能够替他运营这家公司吗？ — Jon Xu > [04:41](https://www.youtube.com/watch?v=R56RJFZBasQ&t=281s) 正确的节奏是构建一个紧密的闭环：深度理解客户需求 $\rightarrow$ 交付产品 $ ightarrow$ 更深入理解需求 $ ightarrow$ 迭代产品。 — Jon Xu > [06:18](https://www.youtube.com/watch?v=R56RJFZBasQ&t=378s) 想法必须建立在“当前模型能力的前沿边缘”（edge of what models can do today）。这意味着产品在现有的前沿模型下可能仅能勉强运行，但随着底层模型的升级，产品效果会呈指数级提升。 — Jon Xu > [06:18](https://www.youtube.com/watch?v=R56RJFZBasQ&t=378s) 在 AI 时代，软件的生产成本正在趋近于零，单纯的工具软件价值大幅缩水，真正的核心价值转向了客户信任、牌照、合规许可以及对结果的掌控。 — Jon Xu > [08:37](https://www.youtube.com/watch?v=R56RJFZBasQ&t=517s) 追求一个野心勃勃的想法与追求一个平庸的想法，所付出的艰辛和时间成本几乎是一样的。 — Jon Xu > [09:13](https://www.youtube.com/watch?v=R56RJFZBasQ&t=553s) 绝大多数创始人刚起步时关注的只是表层的痛点，而真正的商业机会几乎都隐藏在更深层的结构性难题中。深耕的意义并不在于验证最初的那个雏形，而在于“挖掘出藏在底层更好的想法”。 — Jon Xu > [10:27](https://www.youtube.com/watch?v=R56RJFZBasQ&t=627s) 创业中最糟糕的失败模式不是犯错，而是“不作决定”——在多个想法之间摇摆不定，却从未在任何一个方向上深度扎根。 — Jon Xu ## Mentioned Jon Xu, Y Combinator, Blake Scholl, Boom Supersonic, Amazon, Groupon, GovDash, Paul Graham, Corgi Insurance [Source](https://www.youtube.com/watch?v=R56RJFZBasQ)

Source quotes

[0:00] Jon Xu: 第一个常见的失败模式，是执念于寻找“完美想法”。创业固然充满挑战，但完美想法绝对无法通过抽象的空想得来；创业者必须与现实世界真实接触，从客户那里获取反馈，才能明确方向。 [0:00] Jon Xu: 挑选一个充满好奇心的领域并极速深入，短时间内就能积累出惊人的专业知识。 [2:42] Jon Xu: 由于精力和资源被分散，你无法在任何一个项目上深度扎根，从而导致获取的信号和数据极不准确。 [2:42] Jon Xu: 当面对几个看起来同样吸引人的选项时，果断挑选其中一个并全力深耕。所谓“深入”，意味着要“破釜沉舟”（burn the other boats），明确放弃其他备选方案。 [4:41] Jon Xu: 如果明天把你扔进客户的公司，你能够替他运营这家公司吗？ [4:41] Jon Xu: 正确的节奏是构建一个紧密的闭环：深度理解客户需求 $\rightarrow$ 交付产品 $ ightarrow$ 更深入理解需求 $ ightarrow$ 迭代产品。 [6:18] Jon Xu: 想法必须建立在“当前模型能力的前沿边缘”（edge of what models can do today）。这意味着产品在现有的前沿模型下可能仅能勉强运行，但随着底层模型的升级，产品效果会呈指数级提升。 [6:18] Jon Xu: 在 AI 时代，软件的生产成本正在趋近于零，单纯的工具软件价值大幅缩水，真正的核心价值转向了客户信任、牌照、合规许可以及对结果的掌控。 [8:37] Jon Xu: 追求一个野心勃勃的想法与追求一个平庸的想法，所付出的艰辛和时间成本几乎是一样的。 [9:13] Jon Xu: 绝大多数创始人刚起步时关注的只是表层的痛点，而真正的商业机会几乎都隐藏在更深层的结构性难题中。深耕的意义并不在于验证最初的那个雏形，而在于“挖掘出藏在底层更好的想法”。 [10:27] Jon Xu: 创业中最糟糕的失败模式不是犯错，而是“不作决定”——在多个想法之间摇摆不定，却从未在任何一个方向上深度扎根。

Source quotes

[0:00] Jon Xu: You can only figure out what you should be working on by making contact with reality and getting feedback from customers. [2:42] Jon Xu: Going deep... should feel like wearing a new skin. You should become an almost unrecognizable version of yourself. [4:51] Jon Xu: Could you actually run your customer's business tomorrow? [6:18] Jon Xu: In the AI era, the cost of producing software is going to zero... The things that actually become valuable aren't just software for X. They're customer trust, licenses, regulatory permission, and outcome ownership.

Original transcript

[0:01] [music] [0:09] >> Hi, I'm John and I'm a partner at YC. I [0:12] often meet founders who have lots of [0:14] ideas about what to work on and can't [0:16] decide between them. [0:17] Sometimes they're working on multiple [0:19] things. [0:20] Often they'll say that they're waiting [0:22] to find the best idea before fully [0:24] committing, but it's extremely hard to [0:26] make meaningful progress on a startup [0:29] without committing to a single idea. [0:32] So in this video, I'm going to give you [0:35] a rubric for how to stop overthinking, [0:37] pick an idea, commit to it, and then [0:41] figure out fast whether it's actually [0:43] working. The most important piece of [0:45] advice I'd give to founders struggling [0:47] to pick a startup idea is [0:49] don't overthink it. Overthinking a [0:51] startup in the earliest days can take [0:54] many forms, but here are a couple of the [0:56] most common failure modes I see. The [0:58] first is thinking that you need to find [1:00] the perfect idea. In some ways, this is [1:03] an understandable impulse. [1:05] Startups are hard, so shouldn't you [1:07] figure out the best idea before [1:09] committing? The problem with this [1:10] approach is that it's impossible to [1:12] figure out the perfect idea in the [1:15] abstract. You can only figure out what [1:17] you should be working on by making [1:19] contact with reality and getting [1:22] feedback from customers. The second [1:24] overthink is am I the perfect founder [1:27] for this? [1:28] It's true that founder market fit [1:29] matters. A non-technical founder likely [1:32] won't be the right person to come up [1:34] with a killer dev tool startup idea, for [1:36] example. But often founders, especially [1:39] second-time founders, weaponize this [1:41] line against themselves. They convince [1:44] themselves that they need a decade of [1:46] domain experience before they can start. [1:48] The truth is, you don't. If you pick an [1:51] idea you're curious about, go extremely [1:53] deep, [1:55] and most importantly talk to customers. [1:57] It's often possible to develop [1:59] extraordinary knowledge in a short [2:01] amount of time. We see incredible [2:03] examples of this all the time at YC. [2:06] Take Blake Scholl, the CEO of Boom [2:09] Supersonic. Blake spent his early career [2:12] working on ad tech at companies like [2:14] Amazon and Groupon before deciding to [2:17] work on commercializing supersonic [2:20] flight. Lots of people probably thought [2:22] he was crazy. [2:23] But now Boom is a billion-dollar [2:25] company. So don't let the question of [2:27] whether you're allowed to work on [2:29] something stop you from starting. Once [2:31] you've stopped overthinking your ideas, [2:34] it's important to commit to just one. [2:37] Often, I meet founders who are working [2:39] on multiple ideas at once because they [2:41] believe that this is the best way to [2:43] figure out which one will actually work. [2:45] There are a couple problems with this [2:46] approach. The most serious is that it [2:49] tends to produce bad data. If you don't [2:52] actually go deep on an idea, but instead [2:55] juggle it with several others, you won't [2:57] get good signal about whether what [3:00] you're doing actually works. And if you [3:02] don't get good signal, then you could [3:04] either prematurely talk yourself out of [3:06] a good idea or convince yourself that a [3:09] bad one is worth continuing. The [3:11] solution to this [3:13] is to go in depth first. If you're [3:15] trying to decide between several ideas, [3:17] all of which look equally attractive, [3:20] pick one idea and go deep on it. What do [3:22] I mean by going deep? [3:24] The first thing is that you should burn [3:26] the other boats. That is, you should [3:28] explicitly foreclose your other startup [3:31] idea options. Stop working on them. Tell [3:34] any customers that you've pivoted and [3:37] work with single-minded focus on the [3:39] idea you've chosen. One way to think [3:41] about going deep is that it should feel [3:43] like wearing a new skin. [3:45] You should become an almost [3:46] unrecognizable version of yourself. This [3:50] could mean changing your company's name, [3:52] your emails, your website, and even your [3:55] internal narrative about why you're [3:57] building a startup in the first place. [3:58] For example, I worked with a startup [4:01] called GovDash that helps customers win [4:03] government contracts. They pivoted at [4:06] least five times before finding this [4:08] idea. And each time they explored [4:10] something new, they changed their [4:12] company name and how they talked about [4:14] their mission. At one point, I forgot [4:17] how to get in touch with them because [4:18] they changed their email addresses with [4:20] each pivot. By truly becoming domain [4:23] experts in government procurement, their [4:25] fifth idea worked so well that they [4:27] could barely keep up with demand. They [4:29] recently raised a Series B to scale the [4:32] business and meet that demand. Once [4:34] you've decided to fully commit to an [4:36] idea and go deep, how do you know if [4:38] you're actually doing it well? [4:40] The high watermark I use to help [4:42] founders answer this question is could [4:44] you actually run your customer's [4:46] business? Say you want to build voice [4:48] customer service agents for cleaning [4:50] services. The question isn't just [4:52] whether you talked to 20 owners. The [4:54] question is if I dropped you into a [4:56] cleaning business tomorrow, would you [4:59] know how to run it? Do you know what [5:02] their daily crises are? Do you know [5:04] whether answering the phone is a top [5:06] five problem? [5:08] Do you know how much business they lose [5:10] when a call goes unanswered and what [5:12] they would actually pay to never lose [5:14] another one? These are the kinds of [5:16] questions you need to be able to answer [5:18] with very high confidence. [5:20] Another way to think about this is could [5:22] you teach a class on the problem you're [5:24] solving? Are you one of the most [5:26] informed people in the world on the [5:28] subject? Getting to this level will [5:30] involve lots of conversations with [5:32] customers and sometimes even literally [5:35] doing the job yourself. [5:37] But don't obsess over needing to talk to [5:39] hundreds of customers before writing [5:41] code. [5:42] The goal is to do both at the same time [5:45] in a tight loop. Deep understanding of [5:48] customer needs, then product delivery, [5:50] then deeper understanding of customer [5:52] needs, [5:53] then better product delivery. Real [5:55] customers using your product produces [5:57] concrete data that will complement your [6:00] abstract knowledge, giving you a sense [6:02] of whether what you're building is [6:04] actually working. Once you're going deep [6:07] on an idea, there's several ways to [6:09] validate whether it's worth continuing [6:11] to work on. The most obvious one is pull [6:14] from customers, but there's several [6:16] other qualities of good ideas in the AI [6:18] era that you should look out for as you [6:21] go. The first is that the idea sits at [6:24] the edge of what models can do today. [6:27] This might mean that your product barely [6:29] works on today's frontier models, but [6:31] will clearly improve as they get better. [6:34] You should understand the bottlenecks [6:36] impeding your product's performance [6:38] intimately. If a particular bottleneck [6:41] doesn't clear the way you hoped, [6:43] solving that might become the company. [6:46] This is a version of Paul Graham's [6:47] well-known quote that you should live in [6:49] the future and then build what's [6:51] missing. The second quality of a good [6:53] idea is that it should verticalize. By [6:56] this I mean that it should ultimately [6:58] sell an outcome. [6:59] For example, providing insurance or [7:01] medical care rather than just software. [7:04] In the AI era, the cost of producing [7:07] software is going to zero. So, the [7:09] things that actually become valuable [7:11] aren't just software for X. They're [7:14] customer trust, licenses, regulatory [7:18] permission, and outcome ownership. So, [7:20] if you want to get into the insurance [7:22] space, don't build software for [7:24] insurance companies, just be the [7:26] insurer. Similarly, rather than selling [7:29] back office software for banks, just be [7:31] the bank. One example of this is Corgi [7:34] Insurance, an AI-powered commercial [7:36] insurance company from YC's Summer 24 [7:39] batch. They They not content with being [7:41] a tech-enabled broker or even a managing [7:44] general agent because that was just [7:46] owning a part of the solution. Instead, [7:49] they set an ambitious goal of owning [7:51] everything from underwriting to [7:53] providing customer service, the entire [7:55] commercial insurance stack, and even [7:57] took the unprecedented step of acquiring [8:00] an insurance carrier during their YC [8:03] batch to make it happen. Being the [8:04] full-stack insurance company allows [8:07] Corgi to underwrite any insurance line [8:09] in any vertical with a fraction of the [8:12] head count of traditional carriers. [8:15] They can offer far better pricing, much [8:17] faster turnaround, and own all of the [8:20] economics. That brings me to the third [8:22] quality of a good idea. [8:24] It should be the most ambitious version [8:26] of itself. It may seem unintuitive, but [8:29] the cost of pursuing a wildly ambitious [8:32] startup idea and the cost of pursuing a [8:34] modest one are roughly the same. [8:37] They're both extremely hard. They both [8:39] place extreme demands on your time. So, [8:42] aim at the version that, if it works, [8:45] rewrites a sector of the economy. [8:48] Because that's also the version that [8:49] protects you from competitors, attracts [8:52] the best talent, and has a moat worth [8:54] building. This could mean building and [8:56] selling into the most regulated [8:58] industries, like legal, healthcare, or [9:01] financial services, or taking on very [9:04] large incumbents, like a $10 billion [9:06] legacy SaaS company, or building hard [9:09] tech, like robotics for space assembly. [9:12] Now, what if you do all of this and the [9:15] idea fails? The good news is that you'll [9:18] be in a dramatically better position [9:20] than where you started. First, you have [9:23] unambiguous customer data. You know [9:26] whether there's actually a hair-on-fire [9:27] problem in this space or whether you [9:30] just talked yourself into thinking there [9:31] was. You'll have real conviction to base [9:35] a pivot on [9:36] and a better sense of how to execute [9:38] going forward. But more importantly, you [9:41] will often come away from the process [9:43] with a new idea that will actually work. [9:46] When most founders begin, they're [9:47] solving surface-level pain points. The [9:50] real opportunities are almost always the [9:53] deeper structural problems. In other [9:55] words, going deep isn't primarily a [9:57] process for validating the idea you [9:59] started with. [10:01] It's a way to find the better idea [10:02] underneath. This almost always happens, [10:05] especially if you're at the forefront of [10:07] what models can do today. You'll notice [10:10] the bottlenecks, the gaps, the dev tools [10:13] nobody's built. And one of those could [10:16] turn out to be the actual company. [10:18] Here's what I want you to take away from [10:20] this video. First, stop trying to find [10:22] the perfect idea. [10:24] Just pick one. Then, burn the other [10:26] boats. [10:27] Learn everything you can about the [10:28] customer and try to execute for them. In [10:31] the early idea fog, where you can only [10:34] see 10 ft in front of you, the [10:36] temptation is to take a few cautious [10:38] steps in every direction. Sample a [10:40] little here, a little there, stay close [10:43] to home. The problem is that gives you [10:45] almost no information. What actually [10:47] works is to commit to one direction and [10:50] walk fast. You're not guaranteed to end [10:53] up in the right place, but you generate [10:55] much more information per unit of time. [10:58] And when you're walking, you might [11:00] arrive at a better destination, [11:02] one you couldn't have seen from the [11:04] start. The worst failure mode isn't [11:07] being wrong. It's not making a decision. [11:10] Spinning your wheels, [11:12] dabbling between ideas and never going [11:14] deep enough on any one of them to learn [11:16] anything. So pick one and go deep. [11:19] Thanks for watching. [11:21] >> [music] [11:27] [music]