Retell AI创8000万美元年营收 李正韬凭深度交付对抗OpenAI

Retell AI创8000万美元年营收 李正韬凭深度交付对抗OpenAI 面对巨头竞争，语音AI初创团队无需盲目扩招或拼通用基座，而应聚焦大厂边缘化的垂直场景，通过极速执行力与最后一公里定制交付构建生产级壁垒。 Todd Li 详细讲述了自己从 Google 离职并凭借强悍工程执行力打磨业余项目、最终通过 YC 孵化创立 Retell AI 的过程。面对大厂竞争，公司避开通用大模型战场，专注于毫秒级低延迟的实时语音 AI 呼叫中心，以深度的工程交付与合规能力拿下 CVS 等头部企业客户，在两年内达成 8,000 万美元 ARR。访谈中他拆解了早期靠 5 位联合创始人平分股权起步、善用云厂商免费算力、抵制盲目招人与资本毒药等底层实战法则。最后，他分享了以“带薪试工”替代刷题的务实招聘哲学，以及通过顶级薪酬福利确保全力拼搏的员工共享企业成功、实现财富自由的初心。 语音 AI 初创 Retell AI 在成立两年内实现 8,000 万美元年经常性收入（ARR），覆盖超 3,000 家付费客户，并成功打入 CVS、联想、OpenTable 等大型企业的生产级客服场景。 (0:35) 在 ARR 达到 300 万美元之前，团队仅靠 5 位联合创始人运转且全员平分股权；创始人强调必须亲自在一线跑通 0 到 1 验证业务，警惕过早扩招和资金毒药掩盖产品缺陷。 (30:45) 初创对抗大厂策略主打“一米宽、一公里深”的垂直技术纵深，通过前置部署工程师（FDE）深度整合企业业务流，依靠低延迟抗打断的生产级稳定性击败同类竞品。 (20:41) 团队摒弃传统算法刷题，推行带薪实地试工一周的招聘机制，并提供超越 Google 与 Meta 的顶配薪酬福利，承诺以全员财富自由为公司奋斗终极目标。 (60:03)

Retell AI 联合创始人李正韬指出，初创公司在通用大模型时代的制胜关键在于将极窄垂直赛道做到极致，并依靠高密度的执行力与克制的人力扩张，将技术从演示转化为企业级交付闭环。 执行力为唯一杠杆与防范资源陷阱 (15:45) 李正韬提出创业公式为“资源×执行力×运气”，对于零起步团队，唯有执行力完全可控。他认为硅谷并不匮乏资本与网络资源，早期融资过剩或盲目扩充团队反而是“毒药”——若创始人自身认知未达标，便无法考核与驾驭顶级人才。Retell AI 在达到 300 万美元年经常性收入（ARR）前仅由 5 位联合创始人运转，依靠约 100 万美元的各大云厂商算力补贴与极低开支度过孵化期。 在组织治理上，团队采纳 YC 建议采取五人平分股权的架构，杜绝长期利益分歧。日常 99% 的事务由负责销售、工程、产品和运营的对应负责人独立决策，仅有公司战略级转向才需五人一致同意。李正韬强调，创始人必须亲自在业务泥坑中跑通从 0 到 1 的流程，才能形成对交付质量的有效把控。 从浅层工具收敛至企业级语音客服 (18:10) 在经历托福作文批改、直播实时翻译等多次试错被拒后，团队进入 YC 并持续推演方向。因意识到“去口音”与纯文本转语音（TTS）等中间层技术终会被完全替代人工的终局淘汰，加上早期用户一句“为何不直接让 AI 代替我讲话”的反馈，团队在一周内迅速转型为全托管式 AI 语音客服平台。 产品发布后通过 Hacker News 与社交网络获得病毒式传播，并确立了按通话时长计费的用量商业模式。但团队并未停留在小微客户的狂欢中，而是主动选择推进周期漫长、流程苛刻的大企业客户。通过落地办公区物理门禁、设备防火墙与 SOC 2 强合规改造，Retell AI 成功切入 CVS、联想、安克创新等大型客户，以此建立远高于中小客户的净留存率（NDR）与商业壁垒。 纵深工程跨越 Demo 到生产的鸿沟 (20:41) 面对 OpenAI 等科技巨头的潜在挤压，李正韬认为初创公司的破局点在于业务优先级的错位。大厂的垂直业务往往仅由个别总监带领的非核心团队负责，而初创公司则是全员压上。文字客服对系统延迟容忍度可达数秒，但实时语音交互产生数百毫秒的停顿就会导致体验破损，具有极高的工程纵深。 李正韬强调，行业内充斥着表面流畅的 Demo，但能支撑超高并发、处理真实通话插话与口音，并无缝对接企业内部历史数据库的系统屈指可数。为跨越“从 Demo 到生产”的鸿沟，Retell AI 组建并持续扩张现场部署工程师（FDE）团队，深入客户现场搭建专属业务流，以一站式底座加定制化交付的混合模式应对采购竞标。 实战导向选拔与全员财富自由承诺 (60:03) 在招聘上，Retell AI 彻底弃用传统的 LeetCode 算法刷题考核，转为考察真实工程问题解决能力与原始学习速度，并针对高潜力候选人推行为期一周的带薪实地试工，以验证代码习惯与文化契合度。目前公司团队包含 50 名正式员工和 50 名外包人员，其中招聘人员占比极高。 为了匹配高强度工作与严苛筛选，公司提供高于 Meta 和 Google 的市场前 5% 薪酬，以及每日 70 美元餐饮额度、全额医保等多项福利。李正韬表示，团队的终极目标是在公司走向 IPO 或流动性退出时，让全力拼搏的全体成员实现真正的财务自由，否则在道德与机制上均属“根本性的不公平”。 Retell AI 当前年营收已达约 8,000 万美元，并计划将利润反哺于统一真人坐席与垂直电话 CRM 的全链路布局，以及核心模型的后训练。然而，随着基础模型厂商持续优化原生多模态实时交互，初创团队凭借工程集成构建的护城河是否能长期抵御大模型能力的下沉，仍是该模式面临的长远考验。

[13:43] Do you have any other ideas? [13:43] Michael Seibel: This idea doesn't work. But you've built the product well, and the team is strong. Would you be willing to join YC first, then consider pivoting to another idea? [28:37] Why do I still have to speak, translate it, and dub it in real time? Why not have the AI speak for me? [37:59] Michael Seibel: hire as few people as possible [38:43] Todd Li: hard but right thing [40:31] Y Combinator: if you won't split equity equally with someone, why would you call them a co-founder? [42:42] Todd Li: I'm the smartest. I'm the hardest-working. I'm the best [42:42] Todd Li: if you made me do sales, I'd be miserable. [48:16] Early users: never experienced anything like this [48:16] Commenters: How does it sound so much like a real person? [48:16] Todd Li: If the product isn't useful, even if everyone in the world sees it, no one will stick around. [54:34] Todd Li: Speed isn't even the most important thing anymore. Doing something well is what matters most. Sometimes, when you do it properly, you actually move faster. Going fast can be slow, and going slow can be fast. [63:00] Todd Li: 大家已经付出了极大的心血与努力。如果大家这么拼命帮公司做成了，自己最后却没有实现财务自由，这绝对是不公平的。

[0:00] We want everyone here to achieve financial freedom. [0:01] That's our goal. [0:04] Everyone is already working incredibly hard. [0:05] If you work this hard, [0:06] but still can't achieve financial freedom after succeeding, [0:08] I think that would be [0:10] fundamentally unfair. [0:11] Fundamentally, even if we go [0:12] head-to-head with OpenAI, [0:14] we're only competing with one of their directors. [0:16] In other words, it's our entire [0:17] company against one director there. But [0:19] when it comes to [0:21] that person's dedication, [0:22] and their overall understanding, there's no way [0:25] it can match ours. [0:28] Our mentor told us that, to some [0:30] degree, too many customers and too much money [0:33] can be toxic. [0:34] Interviewing 100 fascinating people in Silicon Valley. [0:36] This is Uncle Moon. [0:38] Today we're joined by Todd Li, [0:39] co-founder of Retell AI, whose team joined YC. [0:42] In just two years, the company has reached [0:44] $80 million in revenue. [0:46] This is their office. [0:47] Let's start with a room tour. [0:49] Welcome to our office, everyone. [0:51] How much does it cost each month [0:53] to run a two-story office like this? [0:55] About $50,000 a month, [0:56] including utilities. [0:57] The company has grown quite a bit. [0:59] How many people do you have now? [0:59] We have 50 people now. Right. [1:01] This is our engineering team. [1:02] The engineering team. [1:04] And this is go-to-market, [1:06] our commercial team, [1:07] along with operations. [1:08] They handle company operations. [1:09] That's it for our quick room tour. [1:12] Now let's move on to the in-depth interview. [1:16] Let's begin today's in-depth interview. [1:17] Before we talk about Retell AI, [1:18] let's talk about your earlier experience. [1:21] In 2022, [1:22] you moved to Google's ads team. [1:24] What exactly did you do at Google? [1:26] I mainly developed advertising algorithms. [1:30] Let's say [1:31] Google has a huge amount of traffic. [1:32] how do you allocate that traffic [1:34] so that you can show [1:36] the most relevant ads? [1:37] For example, you want to [1:39] help customers as much as possible, [1:41] while also considering how much revenue it can generate. [1:43] That's roughly what I did. [1:44] So you were optimizing that whole traffic pool. [1:47] Did your time at Google [1:49] help you later as a founder? [1:51] I think it helped a lot. [1:52] Some people [1:55] start a company right after graduation, but [1:57] from my own experience, [1:58] both of my previous jobs [2:00] helped me tremendously. [2:01] At Fortinet, for example, [2:03] my development skills [2:05] improved by orders of magnitude from when I graduated. [2:07] And then, [2:08] at Google, [2:10] its overall system, [2:11] including the culture [2:13] and its standards for code quality, [2:15] showed me how to take a [2:16] product to the highest possible level. [2:18] All of that helped me a lot. [2:21] When did you first want to start a company? [2:22] And when did you begin acting on it? [2:24] Honestly, I think almost every [2:25] software engineer working in Silicon Valley [2:27] has thought about starting a company. [2:29] I was no exception. [2:31] Right after graduating, [2:32] I already wanted to start something. [2:34] I tried many different projects. [2:37] While [2:38] I was at Fortinet and Google, [2:39] I worked on projects of my own, [2:41] and nearly all of them failed. [2:42] But I also gained a lot of experience. [2:44] Looking back now, [2:45] all of those experiences were valuable. [2:48] As for how Retell AI started, [2:50] my [2:52] co-founders and I [2:54] were working on one of those projects, [2:56] and it got into YC. [2:57] That became the catalyst. [2:58] YC requires founders to work full-time, [3:01] so we all decided to accept the offer. [3:03] I think that was [3:04] the real turning point. [3:05] Before that, I had just kept failing. [3:07] So before then, you kept [3:10] trying things in your spare time, [3:12] failing and trying again. [3:14] What people saw later was that you built [3:16] Retell AI and succeeded on the first try, [3:17] but before that, you'd failed countless times. [3:19] Is that right? [3:20] Yes. I think succeeding on the first try is rare. [3:21] Maybe [3:22] some people can do it, [3:23] but from what I've seen, [3:25] most people try many times. [3:27] Before we dive deeper into your founder journey, [3:29] let's talk about another label people give you: [3:32] being a second-generation entrepreneur. [3:34] I really want to use this opportunity to discuss [3:36] what it takes to accomplish something. [3:38] It takes more than just a few resources. [3:41] It requires personal ability, the right moment, [3:43] support from a team, [3:45] and some luck, among other things. [3:47] Only when all of these factors come together [3:49] can you build something meaningful. [3:51] So by taking a closer look [3:54] at your background as a second-generation entrepreneur, [3:57] I hope we can give everyone an honest account [3:59] of why you were ultimately able to succeed. [4:03] Your father [4:03] started a company around the time you were born, [4:05] and later took it public. [4:07] How did his entrepreneurial journey [4:11] help and influence you? [4:13] I think it had many positive effects, [4:14] but also some negative ones. [4:16] On the positive side, [4:18] I learned a lot from his words and example. [4:21] When he was running the company, [4:22] for example, [4:23] when he was in meetings, [4:24] I would listen nearby. [4:25] He would also tell me [4:26] what matters when you're building a company, [4:28] and that it is extremely difficult. [4:31] Most startups fail, for instance. [4:34] Those were [4:34] important lessons with real depth. [4:36] But some things [4:38] may, to some extent, [4:40] have worked against me. [4:44] For example, growing up, [4:45] I probably experienced [4:50] less hardship than most people. [4:52] But entrepreneurship [4:53] is inherently painful. [4:54] So [4:56] endurance [4:57] is extremely important. [4:58] I think my own endurance [4:59] might be a little weaker than other people's. [5:01] If you look at it [5:02] in terms of my upbringing, [5:03] those are the two sides: [5:05] both positive and negative effects. [5:06] That's how I see it. [5:07] You just mentioned that [5:10] he sometimes brought you to meetings. [5:11] Could you tell us more? [5:13] Were there any interesting stories [5:15] or examples you can share? [5:18] I couldn't really understand the conference calls, [5:19] because they dealt with specialized company matters. [5:21] But afterward, he would explain things to me, [5:22] such as what I absolutely needed to watch out for. [5:23] Because he had previously [5:24] encountered certain difficulties, [5:26] or considered certain things important, [5:27] he would tell me about them. [5:29] Also, I grew up in that kind of environment. [5:32] Both my parents worked at the same company. [5:34] They more or less [5:36] started the business together. [5:37] That environment itself [5:39] helped me, [5:40] because it made me believe [5:41] that a startup could succeed. [5:43] Of course, the odds are low, [5:45] but I knew it was possible. [5:46] That gave me much more confidence. [5:49] That was [5:50] a significant shift for me. [5:52] So your whole family are entrepreneurs. [5:55] Your mother is one too. [5:56] When your family [5:57] sat down for a meal, what was the atmosphere like? [5:59] Would you talk about startups, fundraising, customers, and work? [6:02] Were those common topics? [6:03] Yes, very often. [6:05] We talked about those things a lot. [6:06] But growing up, I didn't actually get to have meals [6:08] with my parents that often. [6:10] I went abroad at a fairly young age, [6:11] and most of the time [6:12] I was at boarding school. [6:15] Normally, [6:16] you saw how hard entrepreneurship was for your parents, [6:18] right? [6:18] Yes, extremely hard. [6:20] So when you became a founder yourself, [6:22] weren't you afraid? [6:23] Weren't you worried that once the company truly got going, [6:26] you'd be exhausted every day? [6:28] I honestly didn't think that far ahead. [6:29] I think everyone [6:30] comes to entrepreneurship [6:32] with significant misconceptions. [6:33] For example, I watched Elon Musk's [6:36] interviews, [6:37] as well as Jeff Bezos, Mark Zuckerberg, and others. [6:39] It made entrepreneurship seem [6:40] like it wasn't that difficult. [6:42] But only [6:43] after you experience it yourself [6:44] do you realize how hard it is. [6:45] Also, my parents' company [6:47] went public when I was [6:51] around seven or eight. [6:52] After that, [6:54] they weren't under quite as much pressure. [6:55] What I saw instead were [6:57] the benefits that came with their success, [7:00] such as improvements [7:01] in our life and material circumstances. [7:02] So in my mind, entrepreneurship always [7:04] seemed like something wonderful. [7:05] I see. So [7:06] in reality, it's extremely painful. [7:08] Yes. [7:08] What you saw [7:10] was what came after the IPO, [7:11] including your family's improved financial situation. [7:14] You were old enough to remember that. [7:15] That made it seem as though [7:16] once a startup succeeds, [7:18] it brings nothing but benefits. [7:19] You didn't see [7:20] the dark and difficult parts of entrepreneurship. [7:22] In a way, that may have been a good thing. [7:23] If you'd seen how hard it was, [7:24] you might never have started a company. [7:26] That's true. [7:27] Here in Silicon Valley, [7:28] so many founders are gathered in one place. [7:30] Right. [7:30] There are countless founders. [7:31] Exactly. [7:32] And you see countless successful people. [7:33] Right. [7:34] You can also attend [7:35] all kinds of founder events. [7:37] You can see what happens to people [7:38] after they succeed, [7:39] and before anyone has [7:41] made it, you can support one another. [7:42] I think that's one reason Silicon Valley [7:43] continues to produce startups. [7:46] Later, [7:47] when your family's company went public, [7:49] your father became very successful. [7:54] For your personal growth, [7:56] was that more pressure or more help? [7:58] I think it was more helpful. [8:00] Mainly because [8:01] founders really need mentors. [8:04] Right. [8:05] On my entrepreneurial journey, [8:06] I've had three groups of mentors. [8:08] The first was, of course, my father, [8:09] along with my mother. [8:11] They would tell me [8:11] which things founders [8:13] need to pay attention to. [8:15] We also often discussed [8:18] things like [8:18] which trends were emerging, [8:19] which products might succeed, [8:22] and why they might succeed. [8:23] What ingredients would they need? [8:26] The second group of mentors [8:28] were the people I saw [8:30] on YouTube. [8:31] They shared a lot, [8:34] and displayed certain common traits. [8:36] I later realized [8:37] that I had some of those traits too. [8:38] Whether in personality, [8:41] my field of study, [8:42] or other areas, I could relate to them. [8:43] So I learned a great deal from them. [8:45] The final group included [8:46] our YC mentor, Michael Seibel. [8:48] He is a founder himself, [8:50] and a co-founder of Twitch. [8:52] So [8:52] he taught us a great deal as well. [8:54] When it comes to entrepreneurship, [8:57] I think having mentors is essential, [8:59] because school doesn't teach you these things. [9:01] When you told your father, [9:03] “I'm leaving Google to start a company,” [9:05] what was your parents' first reaction? [9:07] I think they were fairly worried. [9:09] They had been through it themselves [9:10] and knew that starting a company [9:11] is an incredibly risky undertaking. [9:15] But they also knew [9:17] I'd been working on projects before. [9:19] They just didn't think those projects would succeed. [9:21] Right. [9:21] Also, [9:23] they had set aside some startup capital for me. [9:25] They were asking [9:26] whether I would use their money to start the company. [9:28] Because [9:30] when my father started his business, [9:31] my grandmother gave him [9:33] what I think was tens of thousands of yuan. I don't remember exactly. [9:35] That was his startup capital. [9:36] So my parents [9:37] wanted to do the same for me. [9:38] But I think the world has changed. [9:40] There are more and more investors now, [9:42] so founders no longer need their parents' initial capital. [9:45] I think the world today [9:46] increasingly prioritizes ability. [9:50] That's [9:51] my personal impression. [9:52] Let me explain what I mean. [9:54] Resources, [9:57] especially in Silicon Valley, [9:58] are abundant. [10:00] If someone can't start a company [10:02] or make something work, [10:03] it's definitely not because resources are scarce. [10:05] It has more to do with [10:06] one's judgment, abilities, team, and other factors. [10:09] It's a combination of factors. [10:10] You can't attribute your lack of success [10:12] to insufficient resources. [10:13] Is that what you mean? [10:14] Yes. For example, [10:17] imagine Elon Musk [10:19] and an ordinary person [10:22] swapped bodies. [10:23] Suppose Elon Musk [10:24] kept all of his memories. [10:26] In other words, this ordinary person [10:28] would suddenly become extremely capable, [10:29] but would have no [10:30] resources— [10:30] no money, [10:31] or network. [10:33] I believe he could build something huge within a few years. [10:36] Because Silicon Valley [10:37] never lacks money. [10:39] What it lacks are capable founders. [10:42] That has always been the case. [10:44] That's why improving your own abilities matters most. [10:45] I firmly believe that. [10:47] I once watched a show [10:50] about millionaires starting over from zero. [10:52] It took several millionaires [10:53] and billionaires and dropped them somewhere unfamiliar [10:56] to see whether they could start over from nothing. [10:58] I believe one of them succeeded. [11:00] That's why I firmly believe [11:01] successful wealthy people have their own [11:03] logic for building successful companies. [11:04] A successful founder [11:06] can be placed somewhere entirely new [11:07] and still succeed again and again. [11:08] I think money, [11:11] resources, and connections [11:13] are ultimately [11:15] a reflection of a person's abilities. [11:17] They're the product of ability compounded over time. [11:20] Exactly. [11:21] I completely understand. [11:22] Right. [11:23] Watching how you've spoken in this interview, [11:25] yes, [11:26] if people watch more [11:28] Elon Musk interviews, [11:29] they'll notice that Todd's way of expressing himself [11:32] really is very similar to Elon Musk's. [11:35] He's definitely not an extremely extroverted person [11:36] who holds forth on every subject. [11:39] He's genuinely trying to share [11:41] the thoughts from deep inside him. [11:44] Personally, [11:44] I've interviewed many people [11:46] and met many more. [11:48] Sometimes, people who appear very introverted [11:50] and speak thoughtfully, [11:52] unhurriedly, [11:53] even with a few stumbles, [11:57] are incredibly valuable to hear from. [11:58] There is so much [11:59] to discover in what they say. [12:01] Let's continue the interview. [12:03] All right. [12:04] So, [12:05] when I interviewed your father, [12:07] he said [12:08] the greatest challenges in entrepreneurship are [12:10] time, and an opponent who may [12:14] suddenly appear without warning. [12:15] Regarding those two points, [12:16] how did you see them before? [12:17] And now, after two or three years as a founder, [12:19] how do you see them? [12:20] Right. [12:21] I completely agree. [12:22] Entrepreneurship is fundamentally a competitive environment. [12:24] Right? [12:24] You can think of it like being on a soccer team [12:26] playing in the World Cup. [12:29] Basically, only the top one or two teams win. [12:31] That's how it feels to me. [12:33] Another point is, [12:35] when people discuss entrepreneurship, [12:38] they often say [12:38] that you're providing value. [12:40] If you provide that value, [12:42] people will reward you accordingly. [12:43] That's [12:44] the money you can earn. [12:46] Then you can use that reward [12:47] to create even more value. [12:50] But don't forget: [12:52] once large language models arrived, [12:53] everyone could provide that value. [12:55] Yet only the strongest provider [12:56] can truly capture the reward, [12:58] because customers will ask, [12:59] “Who can give me the most value?” [13:00] So fundamentally, [13:02] when everyone can provide value, [13:04] it still comes down to [13:05] who can provide more of it, [13:07] or who has [13:08] something uniquely theirs to offer. [13:09] Time is difficult [13:10] because the market never stops changing. [13:12] Even giants like [13:15] Microsoft and NVIDIA [13:16] continually face [13:17] new technologies [13:18] and new competitors. [13:19] They keep emerging. [13:21] If a company is making money, [13:22] people can see it. [13:24] They'll want to build [13:26] a similar product, [13:27] or create a better one and replace it. [13:29] That's why I think entrepreneurship [13:32] is a road with no end. [13:34] You have to keep climbing incredibly hard, [13:35] but there may always be [13:37] new competitors appearing, [13:39] along with new waves of technology. [13:41] That's true. [13:42] Now let's officially [13:43] dive into the story of building Retell AI. [13:46] According to the official website, [13:47] you left Google in January 2024 [13:50] to work on Retell AI full-time. [13:52] What made [13:53] you feel that you had to leave then? [13:55] You mentioned YC earlier. [13:56] Were there any other reasons? [13:59] We applied to YC twice. [14:00] The first application failed. [14:02] Our first product was [14:04] an app that graded TOEFL essays. [14:05] We did get an interview, [14:08] but the first thing [14:10] our YC interviewer asked was, [14:11] “Do you have any other ideas?” [14:12] They thought our team was good, [14:13] but they thought [14:15] our business direction was completely wrong. [14:16] For our second application, [14:19] we built [14:22] live-stream translation software. [14:23] It had a delay of about eight seconds [14:25] and could translate a Chinese livestream into English. [14:28] During the second interview, [14:29] the interviewer was YC partner Michael Seibel. [14:31] He was already a veteran founder in livestreaming. [14:33] After seeing our idea, he again said, [14:35] “This idea doesn't work. [14:37] But you've built the product well, [14:38] and the team is strong. [14:39] Would you be willing [14:40] to join YC first, then consider [14:42] pivoting to another idea?” [14:44] I think that was the catalyst. [14:45] Another factor was that [14:47] after ChatGPT took off, [14:49] we began [14:51] to research [14:53] the patterns that follow the emergence of new technology. [14:56] Think of the internet, [14:58] mobile technology, or cloud technology. [15:00] After a new technology appears, [15:02] its window of opportunity [15:03] usually lasts only two or three years. [15:06] Most companies that ultimately succeed [15:08] are founded during that window. [15:09] So at the time, we were racing against the clock. [15:11] We wanted to move quickly [15:13] and try different projects [15:14] to see which one could gain traction. [15:16] Then we could build that one out. [15:17] What happened later suggests [15:20] we were probably right. [15:21] Because if we hadn't [15:23] entered the voice AI space, [15:24] there would have been [15:25] other people doing it. [15:26] As for people entering the space now, [15:28] I think they probably won't succeed, [15:30] because there are already [15:31] many competitors. [15:33] Whether startups [15:35] or industry giants, [15:37] everyone is moving into this space. [15:39] So if you enter now, [15:40] you'll have a very hard time winning customers. [15:42] Yes, I think it would be very difficult. [15:44] Right. [15:45] So on the day you quit, [15:46] Retell AI didn't exist yet. [15:48] Correct. [15:49] You'd simply been [15:51] trying countless products in your spare time, [15:52] and all of them had failed. [15:53] But YC liked your team [15:54] and asked you to become full-time founders first, [15:56] then develop the product over time. [15:57] Is that what happened? [15:57] Yes. We had proven [15:59] our team's capabilities. [16:00] Right. [16:01] But we still had neither a product [16:02] nor a market. [16:05] I think the most important thing for a founding team [16:07] is the ability to execute. [16:08] Strategic understanding [16:10] depends heavily on experience. [16:11] That's where mentors can help. [16:12] Whether they're your parents, [16:15] teachers at school, [16:16] or successful founders, [16:18] all of them can offer [16:19] strategic advice and experience. [16:22] But they can't execute for you. [16:24] So for a founding team, [16:26] it's normal to lack insight [16:28] or a good strategy at first. [16:30] The most important thing is still execution. [16:32] I've heard many [16:34] successful founders say [16:35] that startup success equals resources times execution times luck. [16:38] It's the product of all three. [16:41] That's the idea. [16:42] Early in a young founder's journey, [16:45] the only thing they can control is execution. [16:47] If you execute well, [16:49] people will eventually come to help you. [16:50] But if you execute poorly, [16:51] you won't get the other two factors either. [16:53] Exactly. If you execute well, [16:54] it's okay to try things almost at random. [16:55] I remember that at YC, [16:57] there were other teams [16:58] their initial [16:59] ideas didn't work either. [17:00] But I later noticed that teams with strong execution [17:02] and strong technical skills [17:04] could try several ideas in succession, [17:06] and eventually one might take off. [17:07] Along the way, they gain experience [17:09] about what might work [17:10] and what might not. [17:11] On the technical side, [17:12] they also learn [17:13] how to turn technology into a product. [17:15] There's a lot of knowledge involved, [17:17] including go-to-market. [17:19] After a team [17:20] has gone through several experiments, [17:22] it can usually find its own way forward. [17:25] That's why I think [17:26] execution remains the most important thing. [17:27] I completely agree. [17:28] We get many questions from viewers [17:30] who keep asking, “What should I choose?” [17:32] What I want to tell everyone is this: [17:35] what you choose truly isn't that important. [17:37] Just pick something, [17:39] then execute like crazy. Build one thing; if it fails, [17:42] build another. If that fails too, [17:43] keep building. [17:44] Eventually, you'll find [17:46] that one direction works. [17:47] Right. [17:48] Choosing well at the start [17:49] doesn't guarantee success. [17:50] What seems like a good choice now [17:51] is based only on your current understanding [17:53] and today's environment. [17:55] What truly matters is execution, [17:57] which gives you the freshest firsthand information. [18:00] Then you keep adjusting, [18:01] and execution ultimately carries you forward. [18:03] Speed is the foundation of everything. [18:05] Yes, exactly. [18:06] So the five of you [18:08] had already worked together for a long time, [18:10] and YC believed in your team. [18:12] So [18:13] what you thought was, “Okay, [18:14] if YC wants us to become full-time founders, [18:15] we'll do it.” [18:16] That's the full story of how you left your jobs to start a company. [18:18] Once you did, you began experimenting again. [18:21] How long did it take [18:22] to arrive at the idea for Retell AI? [18:24] At that time, [18:26] our mentor would give us [18:27] advice, [18:28] and explain [18:29] how the startup world worked. [18:31] We learned an enormous amount then. [18:32] We'd talk for an hour or two each time. [18:34] We would constantly [18:36] pitch him different ideas, [18:38] all of them related to voice. [18:40] I remember one idea [18:41] was to help people with accents. [18:45] For example, some people [18:46] working in call centers have accents. [18:47] Our product could remove those accents. [18:51] But then we realized [18:54] that “removing accents” [18:56] would soon be a thing of the past. [18:58] If AI eventually eliminates the need for human agents, [19:00] why would anyone still need this product? [19:01] We might as well build [19:03] an AI product that replaces human agents. [19:05] We also had another idea: [19:06] building text-to-speech, or TTS. [19:09] But then again, [19:11] why build TTS? [19:12] TTS is [19:14] technology that turns text into speech, [19:15] in the form of a model. [19:16] Why build it? [19:17] Fundamentally, it's also meant [19:19] to let AI take over [19:22] customer-service work from humans. [19:23] So we might as well go directly [19:25] to the deepest layer, [19:28] the layer closest to the customer. [19:30] That's how we ultimately pivoted to the idea for Retell AI. [19:33] I see. [19:33] Could you explain exactly [19:35] what service Retell AI provides? [19:37] You can think of us as [19:39] a voice-AI customer-service call center. [19:41] That's one way to describe it. [19:41] Take a B2C company, for example. [19:43] It may have millions of customers, [19:45] or even tens or hundreds of millions. [19:48] Such companies commonly run [19:49] a call center with thousands of agents, [19:51] or even tens of thousands, [19:52] answering calls around the clock [19:54] and providing customer service. [19:55] But it isn't [19:59] limited to customer support. [20:01] It also includes sales, [20:03] payment collection, [20:05] and various [20:08] marketing-related use cases. [20:11] We can handle all of those. [20:13] Right. [20:14] Currently, [20:16] our annual revenue is about $80 million. [20:19] We have more than 3,000 customers. [20:21] Some are well-known companies, [20:25] including OpenTable and Opendoor. [20:29] We also serve [20:30] high-profile startup Mercury, [20:32] that's right. [20:33] along with consumer-electronics companies, [20:35] such as Anker, [20:36] Lenovo, [20:40] and CVS. [20:41] In AI customer service, [20:44] OpenAI, [20:45] Anthropic, [20:46] and ElevenLabs are all building products. [20:47] Yet you've still won [20:49] so many major customers. [20:51] What won them over? [20:53] I don't think those industry giants [20:56] are actually doing as well as they appear. [20:57] Here's why. [20:59] Well, [21:00] they work on many different projects, [21:02] whereas voice is [21:04] an extremely deep field. [21:06] It's much harder than text-based customer service. [21:08] With text, [21:09] your [21:10] model, [21:11] knowledge base, [21:12] and every other component [21:14] can have ten seconds or more [21:16] for computation. [21:19] In other words, you have a buffer. [21:20] But with voice, [21:21] all the components, [21:22] including inference, [21:23] the servers, [21:25] and every connection, [21:28] all work differently. [21:29] That's why focus matters so much here. [21:32] Another point is that [21:33] everyone can build an impressive demo. [21:35] They can record an AI phone call, [21:37] post it online, [21:38] and get a positive reaction. [21:39] But going from a demo [21:44] to something a company will accept [21:46] and deploy as a working customer-service system, [21:48] one that truly replaces its existing call center, [21:49] is [21:51] a very long journey. [21:52] And you'll encounter an enormous number of problems. [21:55] We've gone deeper than anyone in this area. [21:57] When customers conduct procurement and due diligence, [21:59] our product consistently comes out ahead. [22:01] After completing their tests, [22:03] they generally conclude that ours is the best. [22:04] That's why they choose us. [22:05] It isn't because [22:08] we have a name as big as OpenAI's. [22:10] Yes [22:11] There's no way we'd have OpenAI's level of brand recognition. [22:13] But after hearing your explanation, [22:14] I can clearly see that [22:15] these companies may appear to offer voice [22:17] or customer-service products, [22:19] but those products may sit within [22:21] a small, overlooked team. [22:23] That team may not get many resources, [22:24] because it doesn't generate much revenue. [22:26] It may not even be [22:27] a top priority. [22:29] So they don't invest much in it, [22:31] and the product isn't as good. [22:33] You, meanwhile, are a small, highly focused team. [22:37] You focus on one narrow point [22:38] and go incredibly deep: [22:40] one meter wide, [22:42] one kilometer deep. [22:43] That's how an entrepreneur once put it. [22:45] By doing that— [22:46] and by being [22:47] the best team in this field— [22:49] you can beat [22:51] those seemingly unbeatable giants. [22:54] That makes perfect sense to me. [22:55] Stories like this in Silicon Valley [22:57] happen all the time. [22:59] An investor once told me [23:01] that a large company only has a few A-teams— [23:03] truly top-tier teams. [23:04] Only a handful. [23:06] You're not competing with the whole company. [23:08] What matters is whether you're up against [23:10] one of its A-teams. [23:12] If the group competing with you [23:14] isn't an A-team in this field, [23:16] and you are, [23:17] success is only a matter of time. [23:19] After hearing your example, [23:20] I feel it confirms once again [23:22] what I sensed from all this information. [23:25] Yes, exactly. [23:26] I think [23:27] it's extremely hard for one company [23:28] to innovate across many different fields, [23:31] because each field [23:32] requires [23:32] incredible depth. [23:34] In practice, if we [23:35] go head-to-head with OpenAI, [23:37] we're only competing with one director's team. [23:39] It's our entire company [23:40] against one team under that director. [23:41] But that team's [23:42] overall commitment to the problem— [23:44] its dedication— [23:45] and its understanding [23:47] could never match [23:49] ours. [23:51] If, however, we built [23:53] the same model as OpenAI [23:55] and competed head-to-head, [23:56] or even entered its second-biggest area, [23:58] coding agents, [24:00] we couldn't beat them. [24:02] Voice just isn't [24:03] that big a priority for them. [24:06] That's fair to say. [24:08] And voice itself [24:09] is an incredibly deep field. [24:10] For example, [24:12] serving our enterprise customers [24:14] is an extremely complex process. [24:16] It can take us— [24:17] after signing a contract—nearly [24:19] a full year to [24:22] fully replace their contact center. [24:24] There are many rounds of testing: [24:26] we keep building and testing, [24:28] validate product stability, [24:31] and work through [24:33] call-related features. [24:35] We also have to migrate [24:37] their existing calling workflows to us. [24:39] All of that takes a long time. [24:42] That reminds me of [24:43] something Naval said: [24:45] in the previous era, development was expensive, [24:48] so companies built SaaS products [24:51] to serve as many businesses as possible. [24:54] Today, development is cheaper, [24:57] so founders need to [24:59] go deep into a company's workflow [25:01] and stay tightly integrated with it. [25:03] They need Forward Deployed Engineers [25:06] to help companies make a deep AI transition. [25:10] That's what creates a moat [25:11] and makes the solution harder [25:13] for competitors to replace quickly. [25:15] Your example [25:17] really deepened my understanding of this model. [25:20] What's your take on that? [25:22] Yes, I agree. [25:23] If it's just [25:25] a highly standardized product, [25:27] then [25:28] it won't be worth as much. [25:31] Most of its features [25:33] may be irrelevant to that customer, [25:35] and a standardized product [25:36] can never go as deep as [25:37] one built specifically [25:39] for that company. [25:40] It simply can't be as powerful. [25:41] What we offer is [25:42] more of a hybrid model. [25:43] Our platform [25:47] can provide every voice-related capability, [25:49] all in one place. [25:52] But the last-mile delivery [25:54] still has to be tailored to each company. [25:56] We help them build the final workflow. [25:59] That reflects a classic YC lesson: [26:02] don't worry about [26:04] scaling from day one. [26:06] First, take your service [26:08] to two or three companies [26:10] and make it exceptional. [26:11] Make them absolutely love it. [26:12] That's a great place for a startup to begin. [26:15] Exactly. At the beginning, [26:17] you don't need much money [26:17] or many people. [26:19] Our mentor told us that, in some ways, [26:21] too many customers and too much money [26:24] can be toxic. [26:27] Take having too many customers. [26:28] What problems can that cause? [26:30] Suppose you're just getting started. [26:31] There are five of you, [26:33] and you somehow have $500 million in the bank. [26:35] Everyone would say, [26:37] having $500 million just sitting [26:39] in the bank isn't a good thing. [26:40] Why collect interest on it? [26:42] Turn that money into people. [26:44] Hire a hundred salespeople, [26:45] a hundred engineers, [26:47] and go build the business. [26:48] But the founder— [26:50] especially a first-time founder— [26:52] probably has no idea [26:54] how to hire great people. [26:55] They'll likely hire the wrong people, [26:57] and those people [26:59] won't build a great product. [27:01] A big part of entrepreneurship [27:03] is developing yourself. [27:06] For a company to be strong, [27:08] the founder has to be its strongest person. [27:10] I don't mean in any one specialty. [27:11] I mean overall. [27:12] Taken as a whole, the founder must be the strongest. [27:14] Only then can they hire [27:17] people who surpass them in their specialties. [27:20] They also have to work the hardest. [27:22] The idea that I can [27:23] sit in the office drinking tea [27:25] while a team does all the work? [27:26] That's impossible. [27:27] At our company, [27:28] we've never seen that. [27:30] The founder always dives in first, [27:33] works like crazy, [27:34] gets the process working, [27:37] then brings someone in to replicate it. [27:39] And whoever takes over [27:40] usually won't do it as well as [27:43] the founder did. [27:44] Quality drops, [27:46] so you have to work hard [27:48] to bring it back up. [27:49] So for a startup, [27:51] the most important question is [27:55] whether the founder can [27:57] level themselves up first. [27:59] It's not about being weak yourself [28:01] and trying to recruit a group of stars. [28:04] Why would those stars work for you? [28:06] You can't hire Jensen Huang to work for you. [28:08] I think that's [28:09] a major misconception. [28:11] Everyone knows people are the most important asset, [28:14] but the most important person [28:15] is still the founder. [28:17] More money [28:19] or more people [28:20] won't change [28:22] or solve the underlying problem. [28:23] If the founder is the problem, [28:25] they have to fix it. [28:27] Like you said, [28:28] Yes [28:28] if the founder's judgment [28:29] or execution hasn't caught up, [28:31] then that company [28:33] will never get off the ground. [28:34] Yes [28:35] Hmm [28:37] That got pretty deep. [28:38] Let's lighten things up a little. [28:41] I read a story about how [28:42] one of your users tried your [28:44] real-time dubbing and translation, [28:47] then asked you, [28:48] "Why do I still have to speak, [28:49] translate it, [28:50] and dub it in real time? [28:52] Why not have the AI [28:53] speak for me?" [28:55] After hearing that feedback, [28:56] you quickly pivoted [28:57] into an AI customer-service company. [28:59] What I'm curious about is this: [29:01] as a company, [29:03] you must get lots of user feedback. [29:06] Why did that one comment [29:08] lead you to pivot so quickly? [29:10] With the products we had before, [29:12] we might spend [29:13] a month marketing them, [29:15] but it was obvious: [29:16] if the market showed no traction, [29:19] we basically got no customers. [29:21] But once we built the new [29:23] voice product, [29:24] we had, essentially, [29:26] tens of thousands of users [29:30] in the first week. [29:33] So if you build [29:35] a product [29:37] that works really well [29:38] and genuinely helps people, [29:40] it should take off quickly. [29:41] Your story really shows [29:42] how a product can go [29:43] from having no users [29:46] for a month or two [29:47] to a successful pivot [29:49] and then [29:51] tens of thousands of users in one week. [29:53] Yes [29:53] That traction [29:55] is the clearest signal, right? [29:57] But how do you judge [29:59] whether to keep going [30:00] for another month or two, [30:01] or give up? [30:02] OpenAI persisted for seven or eight years. [30:04] There are clearly many factors involved. [30:06] There isn't [30:07] a one-size-fits-all answer, right? [30:10] I should take back what I just said. [30:11] OpenAI, [30:12] over those seven or eight years, [30:13] changed direction many times [30:15] and tried many different things. [30:16] They worked on DOTA, for example. [30:19] They must have [30:20] tested many different products. [30:22] I also remember [30:24] Sam Altman talking about ChatGPT. [30:27] They, too, [30:28] didn't know at first [30:28] what product to build. [30:30] There were no coding agents or anything like that. [30:32] They simply released the API. [30:34] It was an interesting piece of technology. [30:36] After releasing the API, [30:37] they still had to decide what product to build. [30:38] They eventually realized a chatbot [30:41] might be the right direction. [30:44] Yes [30:45] So through that whole pivot, [30:46] it was still just the five of you [30:48] trying all those products? [30:50] Yes. We didn't hire our first engineer [30:53] until we reached $3 million in annual revenue. [30:56] So the five of you [30:57] barely paid yourselves during YC [30:59] and just kept working. [31:00] Right. YC gave us $500,000, [31:03] which, for us, [31:04] was already a lot of money. [31:05] If you aren't hiring, [31:06] it's almost impossible to spend it. [31:08] An office costs a few thousand dollars, [31:10] and each person takes a few thousand in salary. [31:12] Then there's food and other basics. [31:13] That money could [31:14] last us several years. [31:17] For servers [31:17] and development costs, [31:18] the cloud providers [31:19] give you lots of credits. [31:20] Amazon gave us [31:23] $300,000 in credits. [31:26] Google Cloud gave us another $300,000, [31:28] and Azure another $300,000. [31:29] So you had nearly $1 million in credits [31:30] to build with. [31:32] They all had [31:33] programs supporting startups. [31:35] That proves the point again: [31:36] if you're starting a company in Silicon Valley, [31:38] resources aren't the problem. [31:39] What matters is execution [31:40] and getting your mindset right [31:42] so you can move quickly. [31:44] And there are more resources all the time. [31:45] Beyond YC, [31:47] other funds [31:49] run similar accelerators. [31:51] The basic idea [31:53] is to help more founders [31:56] with the ability [31:57] move faster [31:59] without being held back by resources. [32:01] Silicon Valley [32:02] really scaled up this model. [32:04] Twenty or thirty years ago, [32:05] for example, [32:08] there weren't nearly as many resources. [32:10] There weren't as many investors, [32:11] or cloud credits. [32:13] Then again, there was less competition. [32:15] So I don't know which [32:16] era made entrepreneurship easier. [32:18] As you—or your father in an earlier interview— [32:19] said, [32:22] starting a company in Silicon Valley [32:23] seems to keep getting easier. [32:25] Yes, that's how it looks. [32:27] Let's talk about the YC experience. [32:30] YC gave you $500,000. [32:33] People in Silicon Valley often say [32:35] YC got a lot of equity for that money. [32:38] But the real value wasn't the $500,000. [32:40] It was the incubation [32:42] and all the [32:43] connections and resources. [32:46] What was the biggest benefit [32:48] YC gave you? [32:50] The biggest benefit was [32:52] how it shaped our thinking and strategy. [32:54] As we just discussed, [32:56] the most important thing in a startup [32:58] is improving the founder's own abilities. [33:00] Those abilities have two sides: [33:02] judgment [33:03] and strategy on one side, [33:03] execution on the other. [33:05] When it came to execution, [33:06] even back then, [33:07] we were already very strong. [33:08] We'd worked [33:09] at big tech companies [33:11] and voluntarily kept 996 schedules. [33:12] Those experiences [33:14] had trained our execution muscles well. [33:16] What we lacked [33:18] was experience. [33:20] That's where [33:21] YC helped us the most. [33:23] Personally, I think [33:24] without YC—or, more specifically, [33:26] if we hadn't [33:27] met our mentor— [33:28] we might not have made it. [33:30] Our gap [33:31] in judgment was simply too large. [33:33] We'd come up with ideas, [33:35] and he would kill the bad ones. [33:37] Without him [33:39] killing those ideas, [33:39] we might have needed [33:41] many more attempts before succeeding. [33:44] I've interviewed and spoken with many people [33:46] and some see this very differently. [33:49] They say YC is useless. [33:50] Hmm. [33:50] What do you make of that? [33:52] Whether it's YC specifically doesn't matter. [33:54] But [33:55] you do need a mentor. [33:56] That could be a family member [33:58] or a friend. [33:59] Ideally, though, that person [34:03] has founded a company [34:04] or has achieved some success. [34:06] I remember taking [34:08] some course once. [34:08] The instructor [34:11] had never built a successful startup. [34:12] Frankly, he was pretty average, [34:14] yet he was teaching the class. [34:15] I didn't understand that at the time, [34:16] but now I realize [34:17] the class probably wasn't very useful. [34:19] Right. [34:19] So even outside YC, [34:21] I'd recommend [34:22] finding [34:24] a mentor. [34:25] YC itself isn't the important part. [34:27] Find someone with real results, [34:28] someone who has successfully built a company, [34:31] and learn from them. [34:32] The question is how to reach them. [34:33] YC provides a channel for that. [34:35] For most people, [34:37] before they've proven themselves, [34:38] it's hard to get access to such mentors. [34:40] But with YC, [34:41] you can apply [34:42] and earn your way in. [34:44] Otherwise, you can knock on doors everywhere, [34:45] but those mentors [34:46] probably won't respond. [34:47] It's like doing a PhD. [34:48] Without [34:49] an advisor, [34:50] the challenge [34:51] would be beyond [34:52] what most people could manage. [34:53] Beyond improving your judgment [34:56] and sharing experience, [34:57] did YC provide concrete resources— [34:59] introductions to customers [35:01] or investors, [35:02] help with recruiting, [35:03] or marketing? [35:04] Things like that. [35:06] The mentor's experience [35:08] was the most important thing. [35:09] Second was what you mentioned: [35:10] marketing support. [35:12] For us, [35:13] once we got into YC, [35:15] many investors [35:17] started reaching out, [35:18] asking to discuss our company. [35:19] Before YC, [35:22] no investor had ever [35:23] wanted to talk to us. [35:24] That change alone [35:25] showed us [35:25] YC's multiplier effect [35:27] and gave us tremendous confidence. [35:29] Another example: [35:30] when we launched our product, [35:33] YC had its own forum, [35:35] a big presence on Twitter, [35:38] and channels such as LinkedIn. [35:41] Portfolio companies [35:43] could use those forums [35:46] to share startup-related news. [35:48] Sometimes, [35:49] all you need is [35:52] a tiny bit [35:53] of fuel. [35:54] Just a little fuel. [35:55] YC is an amplifier. [35:57] It helps the world see [35:58] that you're a strong team. [36:00] You still need to be strong yourself, [36:02] but even a great product needs visibility. [36:04] YC incubates you [36:06] and helps you [36:06] ignite those distribution channels. [36:08] Across all these inputs, [36:10] execution is what matters most. [36:12] Once you can execute, [36:15] many people can help— [36:16] with money, [36:18] advice, [36:19] or marketing. [36:20] But the most important thing [36:21] is still core execution. [36:22] Execution itself [36:24] is difficult, [36:26] but you can practice it [36:28] in many parts of life: [36:29] while studying, [36:31] by trying your own projects, [36:32] or while working at a company. [36:34] Keep raising the bar. [36:37] Ask how you can do the best possible work. [36:38] Treat the company [36:39] as if it were your own and practice executing. [36:42] All of that helps enormously. [36:45] Execution, execution, execution. [36:46] Almost everything you've said today [36:48] comes back to execution. [36:49] I believe in it deeply. [36:51] What do those successful people [36:53] see in you? [36:54] They see your qualities: [36:56] you move quickly, [36:57] you work efficiently, [36:58] and you deliver excellent work. [36:59] If you can do that [37:00] and show those qualities, [37:02] then with some resources, [37:03] some support, [37:05] and a few attempts, [37:06] you're very unlikely [37:08] to fail completely. [37:10] You may not achieve extraordinary success, [37:11] but total failure is unlikely. [37:13] And today, [37:16] these experiments are almost risk-free. [37:18] You can pursue [37:21] small side projects [37:22] outside working hours [37:25] without [37:26] quitting your job. [37:29] You invest time, of course, [37:30] but unlike before, [37:33] you don't need to max out your credit cards [37:35] for a single shot. [37:37] That was [37:39] a major issue [37:40] in the previous era. [37:41] Today, [37:43] that problem is largely gone. [37:44] If you build something in your spare time, [37:46] people can see your passion [37:47] and your ability to execute. [37:48] That's the best proof you can offer. [37:50] Don't deliberately chase [37:52] some so-called benefactor. [37:54] Your execution is your greatest ally. [37:56] I completely agree. [37:59] Have you ever strongly [38:00] disagreed with YC, [38:02] seen things completely differently, [38:03] and struggled to accept their advice? [38:05] Often. We disagreed all the time. [38:07] Tell everyone [38:08] about those huge arguments with your mentor. [38:11] What did those arguments look like? [38:12] What were they about? [38:13] Hiring, for example. [38:15] Yes, hiring. [38:16] He kept telling us to hire as few people as possible. [38:20] Our mentor, [38:21] Michael Seibel, [38:23] often raised points [38:25] that we strongly disagreed with. [38:27] I don't know whether he did it intentionally, [38:29] but he always turned out to be right. [38:31] Hiring was one example. [38:33] Strengthening the founders [38:34] mattered most. [38:35] And to hire well, [38:38] you first have to become highly capable [38:39] so you can attract great people. [38:40] That was one lesson. [38:41] Another was that excess money isn't useful. [38:43] If you have too much money [38:44] but not enough judgment, [38:45] you're like [38:47] the proverbial [38:48] spoiled rich kid— [38:49] exactly— [38:50] holding gold but not knowing how to use it, [38:51] handing bars of it to everyone, [38:54] until [38:55] everyone takes you for a fool [38:56] and comes only to spend your money. [38:58] That's what it feels like. [38:59] Hmm [39:00] Right. [39:00] Another story [39:01] is that he told us very early [39:03] to go enterprise. [39:04] That advice [39:05] helped us enormously. [39:07] With SMBs, [39:08] retention [39:10] and customer size [39:11] tend to be lower. [39:13] One advantage of SMBs is [39:14] that they're more likely to be early adopters. [39:16] But in the end, [39:18] most B2B companies [39:19] still go enterprise. [39:22] Enterprise customers [39:23] are stickier. [39:24] They're harder to land, [39:26] but once you're in, [39:27] retention is stronger, [39:28] and [39:29] net dollar retention is better. [39:31] To clarify, [39:32] SMB means small or medium-sized business, [39:34] while enterprise means large customers. [39:35] YC encourages you [39:38] to use small customers to refine [39:40] the overall shape of the product [39:42] and improve its design and development. [39:44] But ultimately, you should target enterprises. [39:47] Landing one major customer [39:49] can make the company successful. [39:50] Yes, I agree. [39:51] Moving from small customers to enterprises [39:53] is an extremely painful process. [39:54] Small customers [39:55] are usually [39:58] eager to use your product. [39:59] They'll praise it [40:01] and tell you how great you are. [40:03] Enterprise customers, however, [40:03] may barely respond. [40:05] Yet you have to shift your attention [40:07] from small customers to them [40:09] and go plead for their business. [40:10] That goes against human nature. [40:13] It really does. [40:14] And truly, [40:15] as we said, it's hard. [40:17] It is very hard. [40:18] It sounds like YC keeps teaching you [40:19] to do the hard but right thing. [40:21] Yes. Doing the hard but right thing [40:23] goes against human nature. [40:25] Entrepreneurship itself does. [40:27] Everything about it does. [40:28] Yes [40:29] Doing the hard but right thing is deeply unnatural. [40:31] Okay, let's talk about your team. [40:34] You have five co-founders, [40:36] while Anthropic has seven. [40:38] Some statistics suggest [40:43] you shouldn't have too many co-founders, [40:45] that two or three is ideal. [40:47] With five people, [40:48] I don't even know how you divide the roles. [40:50] There are five of us. [40:51] Two handle [40:52] go-to-market— [40:54] sales and marketing. [40:56] Two handle product and engineering, [40:59] while I [41:01] run operations. [41:03] That's the basic split. [41:04] How did the five of you come together? [41:06] Three of us went to the same school, [41:07] USC. [41:08] The other two [41:10] were our colleagues. [41:11] At the time, we may have been [41:13] the YC team with the most co-founders. [41:14] Most teams had only two or three. [41:17] But today, [41:18] I think Silicon Valley as a whole [41:20] is gradually moving toward [41:22] companies with more founders. [41:24] The ecosystem is now mature enough. [41:26] Today, for example, [41:27] companies such as Anthropic, [41:29] or teams around people like Jensen Huang, [41:30] have multiple founders, [41:33] with equity [41:34] split fairly evenly. [41:34] That's becoming common. [41:35] The point [41:38] is to make everyone feel treated fairly. [41:40] That's different from many other places. [41:42] YC has always argued that [41:44] no matter how many founders there are, [41:45] they should split the equity equally. [41:47] YC's website says, [41:48] if you won't split equity equally with someone, [41:50] why would you [41:51] call them a co-founder? [41:53] After three years of building a company, [41:55] how do you view [41:58] that counterintuitive advice? [41:59] I completely agree with it. [42:00] It's also a matter of culture. [42:02] If every other company does it [42:06] and yours doesn't, [42:07] how will the founder with less equity feel? [42:09] Why does everyone else get that deal, [42:10] but not me? [42:12] That creates a problem. [42:13] Here, you're actually working with human nature. [42:15] If one founder owns 10% [42:17] and another owns 40%, [42:19] we have to be realistic about human nature. [42:21] Over time, [42:22] problems will emerge. [42:23] The wider environment matters too. [42:25] If everyone outside the company [42:27] uses the same arrangement as yours, [42:28] it's easier to accept. [42:29] Otherwise, resentment [42:30] and similar feelings [42:31] will gradually build. [42:32] That's inevitable. [42:33] That's how we see it. [42:34] From meeting each other [42:37] and working together at big companies [42:38] to where you are now, [42:40] have you had many conflicts? [42:42] Conflicts? [42:42] Of course we've had them. [42:44] Conflict is normal. [42:46] YC generally [42:47] prefers teams with co-founders. [42:48] If there's only one person [42:50] and no one to balance them, [42:52] many of their decisions [42:53] may be wrong. [42:55] Why do people spend so much time [42:57] arguing in meetings? [42:59] Because the best ideas, [43:02] the highest quality, [43:03] and everything else [43:04] emerge from those debates. [43:06] Human nature makes each person [43:08] want to believe: [43:10] I'm the smartest. [43:11] I'm the hardest-working. [43:12] I'm the best. [43:14] If nobody in the company [43:15] can balance that person, [43:17] they're more likely to go rogue. [43:20] That's what I mean— [43:20] they can go down the wrong path, [43:21] and become impossible to pull back. [43:23] Yes [43:23] Someone needs to correct them [43:27] or provide that balance. It's critical. [43:29] Yes, exactly. [43:30] So you believe [43:32] there should be multiple [43:34] co-founders, [43:35] all with equal equity. [43:37] Yes. And one more thing: [43:38] ideally, they should be friends. [43:39] Hmm. [43:40] If they aren't friends, [43:41] then during an argument [43:44] they may perceive criticism as a personal attack [43:46] or feel deliberately targeted. [43:48] But friends [43:49] already have a foundation of trust. [43:50] They understand [43:51] that the disagreement is about the issue, not the person. [43:54] A common piece of advice says [43:56] you should never start a company with friends, [43:58] because the company may fail [44:00] and you may lose the friendship too. [44:01] That's how it feels. [44:02] So I think [44:04] it's still better to start with friends. [44:04] After you've been around the block long enough... [laughs] [44:08] you realize that trust [44:09] takes a long time to build. [44:10] If, at the start of a company, [44:11] you still have to slowly build trust [44:13] with your own co-founder, [44:15] that company will have a hard time succeeding. [44:17] I agree. [44:17] You may as well bring together people you already trust, [44:19] people you know inside and out, [44:20] and move forward together. [44:23] Right. And there's one more thing: [44:26] ideally, what each person does [44:29] should be something they genuinely enjoy. [44:31] That's important. [44:31] For example, if you made me do sales, [44:33] I'd be miserable. [44:35] Over time, [44:36] that wouldn't be a very good arrangement. [44:38] If someone happens to be [44:39] really interested in one area, [44:41] they can own that area. [44:42] If someone else loves another area, [44:43] they can focus on that. [44:44] That way, everyone [44:45] can operate [44:45] with relative independence. [44:47] In their own area, [44:49] everyone trusts their expertise [44:51] and is willing to step back [44:52] and let them make the decisions. [44:53] At least for matters in that area. [44:55] That makes perfect sense. [44:56] So, as the President, [44:59] what do you think is [45:00] the biggest role you play at the company now? [45:01] What do you spend most of your time on? [45:03] One thing that's unusual about me [45:07] is that I don't like doing the same thing all the time. [45:09] I like working across different departments. [45:11] I'll do one thing today [45:12] and something else tomorrow. [45:13] That preference happens to make [45:15] operations [45:16] a good fit for me. [45:18] For example, we have two co-founders: [45:21] one focuses on product, [45:22] and the other on engineering. [45:23] They're immersed in those areas 24/7, [45:26] always working on them, [45:28] or at least spending most of their time there. [45:29] They go very deep. [45:30] And then [45:31] our other two co-founders— [45:32] one handles marketing, [45:33] and the other handles sales— [45:34] are also deeply focused [45:35] on their respective areas. [45:36] Operations, which I run, [45:38] is made up of many smaller functions: [45:41] finance, recruiting, HR, [45:43] and then things like [45:44] customer success and support, [45:46] legal [45:48] and compliance, [45:49] as well as IT and data. [45:51] You've painted a picture of yourself as [45:52] the company's jack-of-all-trades, [45:54] making sure the entire company [45:55] runs smoothly. [45:59] Right. [45:59] At the beginning, [46:00] all of us were either [46:01] doing marketing [46:02] or engineering. [46:03] So I was writing code at first, too. [46:05] Mm-hmm. [46:05] I built the infrastructure [46:07] for the first few versions of our product, [46:10] including the AWS services, [46:11] and figuring out how to keep [46:14] long-running voice servers [46:16] operating as fast as possible, [46:18] plus CI/CD and things like that. [46:21] Then, little by little, [46:23] as more customers came in, [46:25] collecting payments became an issue. [46:27] So I had to spend some [46:28] time on billing. [46:29] Hiring was another one, [46:31] including finding recruiters. [46:32] That also takes a lot of time. [46:34] As each need appears, [46:35] it gradually grows into a new department. [46:36] And then, when we reached [46:38] around fifty [46:40] to a hundred people, [46:41] IT suddenly became a real need. [46:44] We'd never really thought about it before. [46:45] I couldn't personally go around [46:46] setting up every IT system. [46:48] For example, every employee's computer [46:49] needs access controls, [46:51] a VPN, and so on. [46:52] Those needs [46:54] mean you suddenly have to [46:55] create a new department dedicated to them. [46:58] So, really, [46:59] as a company grows, [47:01] more needs emerge. [47:02] And those new needs [47:04] lead to [47:05] new departments. [47:06] That's one way to think about it. [47:08] It must feel like you're watching [47:11] your company grow up every day. [47:13] What kinds of decisions at the company [47:16] ultimately require your sign-off? [47:18] Do you ever need [47:19] one person to make the final call, [47:21] or do all five of you discuss everything? [47:22] Here's how we do it. [47:23] For smaller matters, [47:25] one person can make the call. [47:27] Like we just discussed, [47:28] everyone needs to [47:29] respect one another's [47:30] expertise in their field, [47:32] the time they've invested, [47:33] and the work they're doing. [47:35] That's where 99% of decisions get made. [47:38] But when [47:39] it's a bigger decision, [47:41] the five of us discuss it together. [47:43] We only move in a direction [47:45] once everyone agrees, [47:47] and then we execute. [47:50] Let's move to the next chapter: [47:51] your growth. [47:54] As you just mentioned, [47:55] once you pivoted to Retell AI, [47:57] you grew very quickly, [47:58] soon reaching $3 million in annual revenue. [48:01] How did you find [48:05] your first few customers, [48:06] and how did you choose them? [48:08] After we launched the product, [48:10] we posted about it [48:11] on the YC forum, [48:13] Hacker News, [48:15] Product Hunt, [48:16] Twitter, and LinkedIn. [48:18] We posted everywhere. [48:19] After that, [48:20] people tried the product and said, [48:22] 'I've never experienced anything like this.' [48:24] They thought it was really interesting, [48:25] and it spread completely organically. [48:26] As word spread, [48:27] some influential people started reposting it. [48:29] Once they did, [48:30] that was the moment [48:31] the product really took off. [48:33] I remember people commenting, [48:34] 'How does it sound so much like a real person?' [48:36] 'This is incredible.' [48:38] I remember Garry Tan reposted it, [48:39] and Paul Graham did too, [48:40] along with [48:41] Aaron Levie and others. [48:42] Wow, all those big names. [48:43] Established Silicon Valley figures [48:48] and the hottest rising stars [48:50] were all reposting you. [48:51] Yes. [48:52] After they reposted it, [48:54] our servers crashed. [48:56] Then we knew what to do next: [48:57] fix the servers. [48:59] People also [49:00] started asking questions— [49:01] why does it work this way, and so on. [49:03] That's how the momentum began to build. [49:05] I think [49:06] marketing [49:07] only works when you have a good product [49:08] worth marketing. [49:10] If the product isn't useful, [49:11] even if everyone in the world sees it, [49:14] no one will stick around. [49:16] I completely understand. [49:18] How much did [49:20] your first paying customer pay, [49:22] and when did they arrive? [49:24] Our pricing is usage-based. [49:26] A customer [49:27] builds a voice agent [49:29] with a particular configuration. [49:30] Once it's configured, [49:31] we charge a set rate per minute, [49:33] based on the length of the call. [49:36] So they simply pay by the minute. [49:37] It's a bit like token-based pricing. [49:40] So your website blew up, [49:41] traffic poured in, [49:43] and then the orders [49:45] and payments started rolling in. [49:46] You began making money naturally. [49:48] After it took off, [49:50] everything was fairly straightforward. [49:53] You could see what needed doing. [49:54] If the servers crashed, [49:56] you added more capacity. [49:57] If customers needed something, [49:58] you built it. [49:59] The hard part was that [50:01] once the voice agent was available, [50:04] people used it for all kinds of things. [50:05] Some people, for example, [50:07] used it to call an AI girlfriend. [50:08] So we had to evaluate [50:12] which use cases would be [50:14] most beneficial to us in the long run. [50:16] We're an open platform, [50:17] which means users [50:18] can build anything they want. [50:20] In other words, [50:21] after we released the technology, [50:24] a lot of people tried it. [50:25] From all that usage, [50:27] we chose a direction [50:27] that we believed had the best odds of success. [50:30] Right. [50:30] Over the long term, [50:31] it was the most valuable direction to pursue. [50:33] That was the idea. [50:34] And which direction was that? [50:35] What did you ultimately focus on? [50:36] Large enterprise customers— [50:37] building AI contact centers for them. [50:40] How long did it take to close your first major customer? [50:43] Well, [50:44] we started with only small customers. [50:46] After those early small customers, [50:47] larger ones gradually came in. [50:49] With a small customer, [50:49] once they started using it, [50:50] if they had a problem, [50:51] we might create a group chat to help them. [50:53] Our support team would assist them [50:55] with after-sales support [50:56] and guidance on how to use the product. [50:59] That's usually enough for small customers. [51:01] Mid-sized customers, however, [51:03] often have more [51:05] compliance [51:06] and legal requirements. [51:08] They may conduct company due diligence [51:11] and ask whether you're SOC 2 compliant. [51:13] Once small customers use you, [51:15] mid-sized customers follow, [51:16] because those small customers provide validation. [51:18] And once mid-sized customers use you, [51:19] large enterprises follow. [51:21] Did you ever have to invest enormous effort [51:22] in winning that first [51:24] major customer—the classic difficult journey? [51:25] Can you share that story? [51:26] CVS was one that took us [51:29] quite a lot of time. [51:30] For context, CVS is [51:32] America's largest pharmacy retailer. [51:33] Right. [51:34] So where did the time go? [51:35] How did you win them over? [51:37] Large enterprises have very high standards, [51:39] especially around compliance. [51:41] For example, to meet compliance requirements, [51:42] we recently installed an access-control system. [51:44] When you start a company, you never imagine [51:46] you'll need to install something like that. [51:48] But compliance standards require it, [51:51] along with things like [51:51] firewalls on employees' computers [51:52] and similar controls. [51:53] Ultimately, it's a process of [51:55] understanding the customer's requirements [51:57] and then meeting them. [51:59] If we aren't willing to meet those requirements, [52:01] then we shouldn't tell [52:03] the customer that we can do the job, [52:05] and they shouldn't become our customer. [52:06] Otherwise, we'd ultimately be hurting them. [52:09] Their data could be leaked, [52:11] or someone who has left the company [52:12] might still have access to our systems. [52:13] That would mean [52:15] betraying the customer's trust. [52:16] You surged to over $80 million. [52:18] What do you think was [52:19] the most important thing you got right? [52:20] Was it simply organic growth? [52:21] Or [52:22] did you face [52:24] many challenges along the way [52:26] and make a lot of good decisions? [52:28] I think we're [52:29] a very pragmatic company. [52:30] For example, compared with product work, [52:32] marketing [52:33] takes at most 10% of our time. [52:35] We spend most of our time [52:37] discussing the product [52:38] or gathering customer feedback [52:40] and refining the experience as much as possible. [52:42] The thing we care about most [52:43] isn't even revenue. [52:44] It's [52:45] how much we genuinely help customers. [52:47] Even with revenue, we ask: [52:49] what kind [52:50] of revenue matters most [52:51] to us [52:52] over the long term? [52:53] Because [52:54] the revenue [52:55] we feel best about earning [52:58] comes from knowing that [53:00] we deliver ten or a hundred times the value [53:03] to the customer. [53:04] That's the kind of value we want to provide. [53:05] That's where most of our time goes. [53:07] As for growth, [53:09] much of it has been organic. [53:10] But as we discussed, [53:13] the process of selling to [53:15] enterprise customers [53:17] has taken a lot of effort [53:18] and a long time. [53:20] We've built that capability gradually. [53:23] At first, [53:24] our sales [53:25] and forward-deployed engineering teams [53:28] had seven people in total. [53:30] We've been steadily expanding, [53:33] and now we're at about ten. [53:35] We still have roughly twenty open roles, [53:37] so we can triple down on this: [53:40] working directly on-site with major customers [53:42] and helping them [53:43] execute the entire integration well. [53:46] Yes, that's actually very difficult. [53:48] Another challenge [53:50] is convincing a large enterprise to use your product. [53:52] That's hard, [53:53] because large enterprises usually work only with [53:55] other large companies. [53:56] When a major enterprise is willing to use [53:59] a product from a much smaller company, [54:02] that's an enormous vote of confidence. [54:04] They're making a major commitment, too. [54:06] So we have a lot of empathy [54:08] for the customer throughout that process. [54:12] They've placed so much trust in us, [54:12] so we have to do everything we can [54:14] to deliver for them. [54:15] That mindset [54:16] runs through the entire company— [54:18] from product [54:20] and R&D to sales, [54:23] and all the way to compliance. [54:25] In everything we do, we want [54:27] to do solid work. [54:28] Speed isn't even [54:30] the most important thing anymore. [54:32] Doing something [54:33] well is what matters most. [54:34] Sometimes, when you do it properly, [54:37] you actually move faster. [54:38] Going fast can be slow, and going slow can be fast. [54:39] That's how it feels. [54:41] Your company's culture [54:43] really is remarkably pragmatic. [54:44] After the market shakes out, [54:47] over the long term, [54:48] it's these pragmatic companies [54:49] that are more likely to endure. [54:51] That's because you're genuinely [54:52] putting the customer first [54:54] and solving their problems. [54:55] Customers naturally want [54:58] to keep working with companies like that. [55:00] Exactly. [55:01] I handle [55:02] procurement, [55:03] so I sometimes buy software. [55:05] And with some of that software, [55:06] the salespeople overpromise like crazy. [55:08] Then, after you buy it, you discover [55:09] they said a feature existed, [55:10] but now it doesn't. [55:11] Some sales tactics are wildly exaggerated. [55:14] They say the product has a feature, [55:15] but it doesn't. [55:16] Or they [55:17] claim something works incredibly well, [55:19] only for it to work poorly in practice. [55:21] Whenever that happens, I think: [55:22] I never want our customers to feel this way. [55:24] Exactly. [55:25] If you do that repeatedly, [55:26] it damages the company's reputation. [55:27] If you're constantly overpromising, [55:28] then in the end, [55:29] you may win some orders this year [55:31] using those tactics, [55:32] but the customer cancels the next month. [55:33] They try it and leave. [55:35] Most [55:36] truly successful companies [55:37] we see [55:38] offer very high-quality products. [55:40] Apple, [55:42] Tesla— [55:43] everyone can see it. [55:44] After buying one product, you want another, [55:46] then an even more premium [55:47] and more impressive one. [55:48] Exactly. [55:49] Fundamentally, [55:51] if you want to build something successful, [55:52] this is how you have to do it. [55:54] Right. [55:55] For a long-term business, [55:56] quality has to come first. [55:58] Chinese companies [55:59] are doing very well now, too. [56:00] Anker, for example, [56:02] is one of our customers. [56:03] Then there are DJI [56:05] and Insta360, whose quality is excellent. [56:08] I even think they're better [56:10] than many American and European brands. [56:12] You can absolutely make that case. [56:14] All right, we've finished [56:15] talking about your growth. [56:16] Now let's move on to your challenges. [56:19] This industry [56:20] seems to be getting hotter and hotter. [56:23] Right. [56:23] So many companies [56:25] have entered the market over the past two years, [56:27] and even this year, [56:28] more keep joining. [56:31] There are many well-known Silicon Valley companies, [56:33] not to mention all the startups— [56:34] we won't name them. [56:35] Their voices sound increasingly natural, [56:36] with lower and lower latency. [56:39] How do you make sure [56:40] that your existing [56:41] customers keep using Retell AI, [56:43] while also attracting new ones? [56:45] Most people trying to build voice AI [56:48] really only have a demo. [56:50] Just a demo. [56:51] We had that three years ago. [56:52] The real challenge [56:53] is going from demo to production. [56:54] That's what's truly difficult. [56:55] Others may advertise that they have the technology, [56:57] but are real customers actually using it? [57:00] Are those customers getting value? [57:01] That [57:03] is much harder. [57:05] I think that's [57:07] our biggest difference from other companies. [57:10] Everyone can say, [57:12] 'We have a voice AI product.' [57:13] But during an RFP, [57:16] the results show that our product is the best, [57:18] so customers often choose us. [57:21] Another company, for example, [57:23] may be much larger overall, [57:25] but its actual capabilities [57:27] in voice may fall short. [57:30] In the short term, [57:32] we see that as a major advantage. [57:35] And over the long term, there are other factors. [57:37] What about the long term? [57:38] Your company is three years old now. [57:40] What is your most important goal [57:41] for the next stage and the long term? [57:42] Over the long term, [57:45] there are three priorities. [57:45] The first is [57:46] our product. [57:47] Customers already consider it the best, [57:50] and we're going to double down on it. [57:53] We'll also expand our product surface. [57:55] Right now, we're only a voice platform, [57:57] but we want to [58:00] expand into areas beyond voice AI. [58:01] Human-to-human calls, for example, [58:03] currently use [58:05] a separate SaaS platform for human agents. [58:06] We'll expand into that area, [58:08] so the entire call [58:09] workflow can run through us, [58:11] instead of customers using us only [58:12] for AI-powered calls, [58:15] then switching [58:16] to other software for human agents. [58:17] We can cover that whole process. [58:19] Another area is customer data. [58:21] Right now, customers [58:22] have their own systems, [58:23] such as Salesforce or HubSpot. [58:25] But we could [58:26] develop a product [58:27] specifically for companies [58:29] whose primary customer channel [58:31] is voice calls. [58:33] We could replace their CRM. [58:35] Those are the product initiatives. [58:37] We're also working on model post-training. [58:39] We're hiring AI [58:40] researchers to do more post-training. [58:43] That, of course, [58:45] is another major initiative. [58:46] The third initiative is [58:48] hiring more salespeople [58:49] and forward-deployed engineers [58:51] to help more large [58:52] customers with deployment. [58:54] Across these three areas, [58:55] we can take the profits [58:58] we've accumulated and reinvest them. [59:01] We believe these three initiatives [59:03] are long-term [59:04] but fundamentally right. [59:06] Revenue [59:07] isn't the core objective. [59:08] If revenue were all I wanted, [59:10] I could buy ads [59:10] or do other things. [59:11] But for us, [59:12] what matters [59:13] is how we can [59:15] deliver more value to customers. [59:16] From a first-principles perspective, [59:18] more customer value [59:20] naturally leads to more revenue. [59:22] But [59:22] that conversion takes time. [59:24] So we only need to [59:25] focus on customer value. [59:27] With revenue, [59:28] say you [59:29] increase customer value, [59:31] but revenue doesn't rise right away. [59:33] In the second year, [59:34] when those customers continue using your product, [59:36] that's when revenue may grow. [59:38] So the payoff [59:40] may take twelve or twenty-four months. [59:42] That's the time horizon. [59:42] I completely agree. [59:43] Many prominent founders [59:45] have said the same thing in interviews: [59:46] if you do the important things right— [59:49] technology, product, and the right marketing— [59:52] you may not see $100 million in revenue this year, [59:56] but you will next year or the year after. [59:58] As you were just saying, [59:59] your medium- and long-term goals [60:01] are extremely clear. [60:03] And you'll need a lot of people. [60:04] You're hiring more than thirty right now. [60:06] On the other hand, [60:08] many people in my audience [60:10] keep telling me [60:11] how hard it is to find a job right now. [60:13] So next, let's [60:14] talk about hiring. [60:16] First of all, [60:17] how do you hire? [60:18] For us, hiring [60:20] is mainly about trying to [60:22] test a candidate's ability— [60:23] their ability to do the actual job— [60:25] rather than [60:26] their ability to solve [60:27] a LeetCode algorithm problem. [60:29] I think that's the most important thing, [60:30] because 99% of the time, people are [60:32] doing the actual work. [60:34] Another thing [60:35] we care about [60:36] is the candidate's [60:37] actual [60:39] learning speed and [60:41] what we call raw talent. [60:43] They may not have much prior experience. [60:46] More experience can be a plus, [60:49] but if someone [60:50] is genuinely very smart, [60:51] they can quickly pick up [60:53] new algorithms, [60:53] new AI [60:54] tools, [60:55] and adapt to this environment. [60:57] So those [60:58] are some of the things [61:00] we prioritize. [61:00] But [61:01] the most important point— [61:02] and I still make this mistake— [61:04] is that we shouldn't assess skills first, [61:06] talent second, [61:07] and culture last. [61:09] It should be the other way around. [61:09] The most important question is [61:11] whether this person finds the mission interesting. [61:13] Is it just a job to them, [61:15] or do they genuinely believe [61:16] the work creates value? [61:17] Can they genuinely [61:19] learn something from the work? [61:20] And do they truly [61:21] believe in the company's culture? [61:24] How do you assess those qualities [61:26] during the hiring process? [61:29] We use several methods. [61:31] One thing [61:32] that sets us apart [61:33] is practical coding. [61:35] We don't use LeetCode at all. [61:36] That way, [61:38] even if someone hasn't [61:39] had time to do extensive interview prep, [61:41] they can still pass the interview. [61:44] That's one difference. [61:45] Another is [61:46] our work-trial process. [61:47] If someone [61:49] doesn't perform especially well in the final interview, [61:50] for example, [61:53] but showed us in another round [61:55] that they're highly capable, [61:57] we can invite them to a work trial. [62:00] They can [62:02] come work at our company, [62:04] perhaps by taking leave from their current job, [62:05] without having to resign, [62:07] and work with us for a week. [62:08] Then both sides can see [62:10] what the other is really like at work. [62:13] Both methods serve the same purpose: [62:14] finding a way to make real job performance [62:17] visible [62:20] in an interview setting. [62:21] Of course, there will always be [62:23] some loss [62:25] and some noise, [62:26] because ultimately, [62:27] you only have a few hours [62:28] to evaluate someone. [62:29] So the process [62:30] inevitably loses some information. [62:32] You just said you have fifty employees, [62:35] but twelve people are recruiting for you full-time. [62:38] Is that right? [62:40] We have fifty full-time employees [62:42] and fifty contractors. [62:44] Mm-hmm. [62:44] Two of our full-time employees [62:46] work in recruiting, [62:47] and [62:48] ten contractors also work in recruiting. [62:49] Did YC tell you to invest [62:52] that much manpower in hiring [62:53] and to take recruiting so seriously? [62:55] Yes. [62:56] Mm-hmm. [62:56] Our screening process [62:57] is honestly a little intimidating. [62:58] Our friends all say, [63:00] 'If you're this selective, [63:01] your pay [63:03] and benefits had better be exceptional.' [63:04] 'Otherwise, [63:05] you're running a sweatshop.' [63:06] Exactly. [63:06] At the same level, we probably pay more [63:08] than Meta or Google, [63:10] for comparable roles. [63:12] And our benefits are better. [63:13] Yes. [63:14] What benefits do you offer? [63:15] Our benefits include [63:16] $70 per person, per day [63:17] in DoorDash credits. [63:19] So, [63:19] whether you're working from home [63:21] or at the office, [63:22] you can order whatever you want. [63:24] We also provide [63:25] $300 a month in commuter benefits. [63:28] That can cover rideshares, [63:29] gas for your own car, [63:30] or insurance— [63:31] even auto insurance. [63:33] Then there's also [63:35] $300 in wellness benefits. [63:37] You can use it for a gym membership, [63:38] an Oura Ring, [63:39] or, for example, [63:42] a sleep-aid subscription. [63:44] Our thinking is [63:45] that employees should be able to put all their [63:48] energy into focused work. [63:51] Apart from rent, [63:52] they shouldn't have to worry [63:53] about the other practical details [63:54] of daily life. Our health insurance, for example, [63:55] includes [63:57] options with 100% coverage. [63:59] So none of those benefits are included [64:00] in the stated compensation package? [64:02] They're all additional benefits? [64:04] Yes, exactly. [64:05] And for salary, we aim for [64:07] the top 5% of the market. [64:09] Got it—so it's very high. [64:10] You offer plenty of both RSUs and cash. [64:12] And our thinking is, [64:14] if the company has a successful exit, [64:16] whether through an IPO [64:17] or another path, [64:18] we want everyone to achieve financial freedom. [64:20] That's our goal. [64:22] Everyone already works extremely hard. [64:24] If they work that hard, [64:25] help make the company succeed, and still aren't financially free, [64:27] I don't think that's fair. [64:30] Everyone should replay this part. [64:32] I think a company [64:34] that's willing to share the upside with its people [64:39] is the kind that can succeed. [64:40] I've always firmly believed that, [64:42] because it gives people real motivation. [64:44] After this long interview, [64:47] I think everyone can see you're that kind of person. [64:48] If I became financially free [64:50] but our employees didn't, [64:51] I'd feel deeply guilty. [64:53] Yes, that would be terrible. [64:54] Everyone created all this value together. [64:55] Today's interview [64:57] has been incredibly information-dense. [64:58] Todd, thank you so much. [65:00] Thank you. Thanks for visiting our office.