Transcriptly

How I Built a $1.3B Startup by Pivoting Fast | Windsurf, Varun Mohan, Co-Founder & CEO

EO21:37Added Aug 31, 2026

This episode features Varun Mohan, co-founder and CEO of Windsurf. In 2021, Varun left his autonomous vehicle job to start a GPU virtualization company. They...

Watch on YouTube →
Contributed by 刘嘉琪

Transcript

Transcript format

  1. Opening

  2. 00:00I think startups are basically like
  3. 00:01getting slapped in the face probably
  4. 00:02over and over again. That's basically
  5. 00:04it. I also, you know, interestingly, I I
  6. 00:06actually like failures a lot. One of the
  7. 00:08things I hate the most is doing
  8. 00:09something and not knowing if it's
  9. 00:11working or not. And it's actually very
  10. 00:12it's very freeing to know when something
  11. 00:14fails because that's very obvious. I
  12. 00:16like when decisions are obvious. Like
  13. 00:17when something fails, it's obvious you
  14. 00:19need to do something new. And the faster
  15. 00:20you fail, actually, the faster you can
  16. 00:22decide to do something new. And I think
  17. 00:23for a company like us, like we're at the
  18. 00:25frontier of the technology, we should be
  19. 00:27imagining we are failing at a lot of
  20. 00:28initiatives. In fact, if everything is
  21. 00:30working, we're operating at less than
  22. 00:31potential of the company. We should be
  23. 00:32failing. Like that probably means we're
  24. 00:34not betting. We're not taking enough
  25. 00:35bets if we're not failing enough. I I
  26. 00:37think it's like a required part of a
  27. 00:38company that you that you fail. I'd say
  28. 00:40like be more humble about your ideas.
  29. 00:42And I wouldn't say that I was arrogant
  30. 00:44at the time, but probably a lot more
  31. 00:46idealistic about our ideas being
  32. 00:48correct. And you need some amount of
  33. 00:49idealism as I said before, but at the
  34. 00:51same time, it's that there's no reason
  35. 00:53why an idea is great if you haven't like
  36. 00:55validated it. be honest with yourselves
  37. 00:57about, you know, the viability of your
  38. 00:59idea and be willing to pivot very
  39. 01:00quickly. I feel like a lot of a lot of
  40. 01:02founders hate the word pivot, but it's
  41. 01:03it's awesome. Like, it's awesome. You
  42. 01:05know, what sucks more is like doing the
  43. 01:07wrong thing and just failing. Even right
  44. 01:08now, I would say one of my regrets is
  45. 01:10probably not doing the pivot 3 months
  46. 01:12earlier.
  47. 01:14Hey, I'm Verun, CEO and co-founder of
  48. 01:16Windsurf. So, Windsurf is an AI powered
  49. 01:18ID uh that provides agentic capabilities
  50. 01:21that enables developers and
  51. 01:22non-developers to build apps really
  52. 01:25quickly and also modify existing
  53. 01:27applications really quickly. A million
  54. 01:28developers have actually used the
  55. 01:30product. We have hundreds of thousands
  56. 01:31of monthly active users on the on the
  57. 01:33product right now and it's growing, you
  58. 01:35know, exponentially faster with time.

  59. Why I Chose Smaller Companies Over Big Tech

  60. 01:37So, I graduated from MIT in 2017 and
  61. 01:41after that I actually worked at an
  62. 01:42autonomous vehicle company in Mountain
  63. 01:44View called Nuro. there sort of got an
  64. 01:46early taste of what deep learning could
  65. 01:48do for other industries. Started the
  66. 01:50company actually in 2021, so it's been 4
  67. 01:52years. At the time, the company didn't
  68. 01:54do anything relating to code AI. It
  69. 01:56actually built out GPU virtualization
  70. 01:58and compiler technology. We did that for
  71. 02:01over a year and a half and were able to
  72. 02:03get to a couple million in revenue. We
  73. 02:04had eight employees at the time. One
  74. 02:06thing changed very very materially in
  75. 02:08that time. In the middle of 2022, GPT
  76. 02:113.5 came out at that point and we
  77. 02:13thought that generative models would
  78. 02:14fundamentally transform many different
  79. 02:16industries and we needed to make kind of
  80. 02:18a pretty big decision at that point to
  81. 02:20pivot the company because we
  82. 02:21fundamentally felt everyone was going to
  83. 02:23run generative models for everything.
  84. 02:24And at that point we believed a lot of
  85. 02:26the value would acrue to companies that
  86. 02:28were applications almost like in the
  87. 02:31early days of the internet the companies
  88. 02:32that proved to be very valuable were
  89. 02:34companies like Google and Amazon and we
  90. 02:36wanted to see what would it be like if
  91. 02:38we could build the next Google or
  92. 02:39Amazon. We were early adopters of a
  93. 02:41product called GitHub Copilot and from
  94. 02:43there we actually built out a product
  95. 02:44called Kodium which was an extension
  96. 02:46that lived in all the major IDEs and we
  97. 02:49were able to get that to over a million
  98. 02:50users as well. Um, and we ended up
  99. 02:53serving some of the world's largest
  100. 02:54enterprises, companies like JP Morgan
  101. 02:56Chase. Very quickly though, sort of
  102. 02:58middle of last year, we realized that we
  103. 03:00needed to control more of the
  104. 03:01experience, especially as these models
  105. 03:02became more and more agentic. And that
  106. 03:04was why we decided to build our own ID,
  107. 03:06which was Windsurf. That has taken off
  108. 03:08very quickly as well.

  109. Killing a $28M Business to Start Over

  110. 03:11[Music]
  111. 03:15It's actually kind of interesting. I
  112. 03:16think MIT is very different than
  113. 03:18Stanford. I think most people don't go
  114. 03:20to MIT to kind of immediately start a
  115. 03:22company. I don't think my aspirations
  116. 03:24were to start a company. I did
  117. 03:25progressively intern at smaller and
  118. 03:27smaller companies. So I first interned
  119. 03:30at LinkedIn, then after that Quora uh
  120. 03:32and then data bricks. Data bricks at
  121. 03:34that time was a very small engineering
  122. 03:35team. They were not even a unicorn at
  123. 03:37that time. But I think I didn't actually
  124. 03:39end up working at any of these companies
  125. 03:40because I think the thing that motivated
  126. 03:42me was can I be a meaningful part of a
  127. 03:44visionary company working with motivated
  128. 03:46people. That's always been what's driven
  129. 03:48me and that's sort of why immediately
  130. 03:49after MIT I decided to go work at an
  131. 03:51autonomous vehicle company. I thought
  132. 03:52that could be the future of where just
  133. 03:55robotics was going to go and I think for
  134. 03:57me and probably the other people at the
  135. 03:59company we want to work on the future of
  136. 04:01technology and that's what truly
  137. 04:02motivates us. I think the motiv is just
  138. 04:05how do you build products in hard
  139. 04:07technology spaces where the technology
  140. 04:09is not there yet. You know I I'll give
  141. 04:10like a quick just some interesting
  142. 04:12numbers. You know, when we started,
  143. 04:14obviously, deep learning was was pretty
  144. 04:16popular. Uh, but the amount of compute
  145. 04:18that these models had access to grew
  146. 04:20exponentially year-over-year. You know,
  147. 04:22just some numbers here. In 2017, the
  148. 04:24number of teraflops on a consumer grade
  149. 04:26GPU was 10. By the end of 2022, it was
  150. 04:28actually 700. So, it's a massive
  151. 04:31increase of in compute that happened.
  152. 04:33And I think what we learned was machine
  153. 04:35learning models were going to get more
  154. 04:37capable very quickly. And you should not
  155. 04:39bet on where the technology is today,
  156. 04:41but where it could be a couple years
  157. 04:42from now, right? And if all you're doing
  158. 04:44is working on products that work today,
  159. 04:46you're going to be quickly irrelevant a
  160. 04:47year from now. And that was like a very
  161. 04:49important lesson that we I think a lot
  162. 04:51of a lot of people that were in the
  163. 04:52autonomous vehicle space kind of
  164. 04:54learned, which is now if you were to
  165. 04:55look at autonomous vehicles, they're
  166. 04:57much more machine learning based than
  167. 04:58they were probably 5 or 6 years ago.
  168. 05:00Originally, we started out the name of
  169. 05:02the company was Exaf Function. Part of
  170. 05:03the reason why we called it exunction
  171. 05:05was our goal was to virtualize GPU
  172. 05:07computations to make it easier to run
  173. 05:09things on GPUs. And the the term exa
  174. 05:11function means we want to run an exa
  175. 05:13number of functions which is 10 to the
  176. 05:1418 functions. I think the biggest
  177. 05:16learning lesson for us was at the time
  178. 05:19was even if we were succeeding by some
  179. 05:21metrics we were making some revenue just
  180. 05:23accepting that hey like the business
  181. 05:26might not be the best business and we
  182. 05:27need to go pivot was a very hard thing
  183. 05:29to do I would say right it's very hard
  184. 05:31when you have like employees you know
  185. 05:33you've already raised some amount of
  186. 05:34funding I think at the time we had
  187. 05:35raised over $28 million of funding to
  188. 05:37basically start from scratch overnight
  189. 05:39uh but one of the learning lessons that
  190. 05:41I've sort of had and probably the other
  191. 05:42people at the company is every time you
  192. 05:44do a pivot you have an opport
  193. 05:45opportunity to maybe 10x the size of the
  194. 05:47company. And uh usually when you have an
  195. 05:48idea that you don't believe in and the
  196. 05:50ceiling is low, it's actually better to
  197. 05:52just scrap the entire thing than to like
  198. 05:53try to incrementally claw and you know
  199. 05:55increase the value of what you're doing
  200. 05:56by like you know 20 30%. It's not going
  201. 05:59to matter in the grand scheme of things.
  202. 06:00So we had gone to a couple million in
  203. 06:02revenue at that time. I think the hard
  204. 06:04part for us was it felt very ad hoc how
  205. 06:07we were adding revenue to the business.
  206. 06:09It also felt like with the advent of
  207. 06:11these generative models, a lot of the
  208. 06:13complexity of running models would
  209. 06:14become commoditized. If everyone is
  210. 06:17going to run transformers, uh the
  211. 06:18transformer model architecture, what is
  212. 06:20the reason for us to have a platform
  213. 06:22that virtualizes arbitrary GPU
  214. 06:24computations? It's not as important
  215. 06:25anymore. So, it's a factor of two
  216. 06:27things. We didn't understand how we
  217. 06:29could 10 or 100x sales, right? I think
  218. 06:31you build a company not to make a couple
  219. 06:33million in revenue, but billions of
  220. 06:35dollars in revenue. So we didn't
  221. 06:36understand how to even get within an
  222. 06:38order of magnitude of that right that
  223. 06:40was one and two we also felt that the
  224. 06:41problem we were solving with the advent
  225. 06:43of the generative models was going to
  226. 06:44get commoditized. So at that point it's
  227. 06:46kind of an easy thing once you believe
  228. 06:47that something is not going to be big
  229. 06:49you have no other option but to change
  230. 06:50your mind and do something new that is
  231. 06:52very painful because you did commit a
  232. 06:54lot of time and a lot of energy and
  233. 06:56passion to one thing. You don't win an
  234. 06:58award for doing the wrong thing for
  235. 07:00longer. I guess me and my co-founder did
  236. 07:01a walk over a weekend and I guess we
  237. 07:03decided over a weekend and we told the
  238. 07:05team on Monday. Um and then everyone
  239. 07:06started working on the new thing
  240. 07:08starting Monday. Uh yeah, I guess we
  241. 07:09just you know one of the core beliefs
  242. 07:11that we have about startups is startups
  243. 07:13can if they're lucky do one thing really
  244. 07:15well. We could not afford to have the
  245. 07:17company believe two things were
  246. 07:18important. So we would need to rip the
  247. 07:20band-aid off and actually make it very
  248. 07:22clear to the rest of the company that
  249. 07:23the new thing was where we thought the
  250. 07:25future was. You know, it's interesting.
  251. 07:26I would we were not worried at all when
  252. 07:28we did the pivot. We had basically
  253. 07:29written off the company is going to be a
  254. 07:31zero and at that point anything is
  255. 07:33greater than zero. Ship whatever you can
  256. 07:35if it sucks you're in the same place
  257. 07:37that you are. If it doesn't suck like
  258. 07:39great awesome only upside just honesty,
  259. 07:42transparency and int like intellectual
  260. 07:44honesty with the team is maybe a core
  261. 07:46tenant of the company. We don't do
  262. 07:48things unless we like truly believe they
  263. 07:50are the right thing to do. even if it
  264. 07:51like fits some narrative that the rest
  265. 07:53of the industry or our investors or
  266. 07:56convenient belief we believe things if
  267. 07:58from first principles they're correct
  268. 08:00and you know I guess one true fact was
  269. 08:03after telling our employees there's a
  270. 08:04real chance that some of them would
  271. 08:06leave but I think like when you build a
  272. 08:08highquality culture where the people are
  273. 08:10are intellectually honest um good people
  274. 08:12don't want to leave if you if you chart
  275. 08:14a path for what the thing that does work
  276. 08:17is right because you know one
  277. 08:18interesting thing about startups is
  278. 08:20people talk a lot about moes. What's the
  279. 08:22mode of a company? But really, what is
  280. 08:23the mode of a company that has 10
  281. 08:25people? You know, if you do have a moat,
  282. 08:27it's very shallow at best, right? The
  283. 08:28number of engineering years that went
  284. 08:30into your product is already very small.
  285. 08:32So, the real moat is you work on the
  286. 08:34right thing with enough people for long
  287. 08:36enough. That's like the only moat a
  288. 08:37startup ultimately has. And that's like
  289. 08:39why you believe in a team. Obviously,
  290. 08:41when a company becomes thousands of
  291. 08:42people, hopefully the moat is not the
  292. 08:44current business. It would be sad if the
  293. 08:45moat was the company can just pivot and
  294. 08:47do the next thing. In that regard, I
  295. 08:49think if you genuinely believe the
  296. 08:51motive of a of a company is betting on
  297. 08:53the right things and and hiring a great
  298. 08:56group of people, then if you are
  299. 08:58intellectually honest with your team and
  300. 08:59tell them we don't believe in the
  301. 09:00strategy we just had, it's should be
  302. 09:02totally fine with everyone.
  303. 09:06This episode is sponsored by Atio, the
  304. 09:08AI native CRM for the next era of
  305. 09:10companies. Connect your email and ATIO
  306. 09:13instantly builds your CRM right before
  307. 09:15your eye with every company, contact,
  308. 09:17and interaction you've ever had enriched
  309. 09:19and organized. That's ATIO. And here's
  310. 09:22what makes it even more game-changing.
  311. 09:24You can build AI powered automations and
  312. 09:26use its research agents to tackle some
  313. 09:28of your most complex business processes,
  314. 09:31freeing you to focus on what matters
  315. 09:32most, building your company. Join
  316. 09:35thousands of companies who are already
  317. 09:37using to power their businesses. Visit
  318. 09:40the link in the description to begin
  319. 09:41your twoe free trial with Atio.

  320. Zero to 1M Users in 2 Months - The Growth Engine

  321. 09:49I think we were early adopters of a
  322. 09:51product like GitHub copilot. So we and
  323. 09:54we were we actually built a lot of
  324. 09:56infrastructure. So at the time we
  325. 09:57thought that was the tip of the iceberg
  326. 09:59of what the technology could actually
  327. 10:00build. We thought a lot more could be
  328. 10:02built in this category. Right? If all
  329. 10:04people were doing was autocompleting, we
  330. 10:06could potentially see a world in which
  331. 10:08entire PRs would get generated. And at
  332. 10:10that time, we were like, okay, the time
  333. 10:12it takes to build applications probably
  334. 10:13has gone down by a double digit percent.
  335. 10:15Why don't we set the goal of the company
  336. 10:16to reduce the time it takes to build
  337. 10:18technology by 99%. It's a very ambitious
  338. 10:20goal, not something we'd be able to do
  339. 10:21in the next year or two. But if we set a
  340. 10:24very ambitious goal, there's probably a
  341. 10:25lot of work to be done in this space. We
  342. 10:27believe that technology is always going
  343. 10:28to get better. We believe the models are
  344. 10:30going to get better. Because of that, we
  345. 10:32just believed there was a lot of
  346. 10:33opportunity here. Because we were also
  347. 10:34an infrastructure company at the time,
  348. 10:36we were able to train our own models and
  349. 10:37run models ourselves, which enabled us
  350. 10:39to build an extension that was entirely
  351. 10:41free in all the major IDs. And uh that
  352. 10:44was able to very quickly get to hundreds
  353. 10:45of thousands of users. It was a
  354. 10:47combination of what our skill set was,
  355. 10:48also a belief of where we thought the
  356. 10:50future was going. I think it took us
  357. 10:52less than 2 months to build the MVP and
  358. 10:54ship it out. I think as a company we're
  359. 10:56probably more of the opinion of if you
  360. 10:58are going to do something hard um it is
  361. 11:00actually important to have intermediate
  362. 11:02goals that you can actually give to
  363. 11:04other people to validate your
  364. 11:05hypothesis. That's maybe something that
  365. 11:07we learned a lot of us at the company
  366. 11:08were previously from autonomous vehicles
  367. 11:10and I think one of the biggest mistakes
  368. 11:12that autonomous vehicle companies might
  369. 11:14have made is maybe to no fault of their
  370. 11:16own was they built a hard technology
  371. 11:19where it's very hard to validate if the
  372. 11:20technology is good in the middle. Right?
  373. 11:22So for us it's always been it is
  374. 11:24actually a feature if you can have a
  375. 11:26very very like visionary goal but have
  376. 11:29very tractable intermediate steps and so
  377. 11:31we actually picked a very tractable
  378. 11:33intermediate step to deploy which was
  379. 11:35can we build a basic VS code extension
  380. 11:37that had autocomplete capabilities uh
  381. 11:39with our own model and that was that was
  382. 11:41the intermediate goal that we had and we
  383. 11:43believed you know maybe it was like a
  384. 11:45little bit of necessity we just pivoted
  385. 11:46the company people were not working on
  386. 11:48anything else there was an existential
  387. 11:50dread what the value of the product was
  388. 11:52so we very quickly built it. I think a
  389. 11:55basic form of product market fit was
  390. 11:56when our inbound for the product from
  391. 11:58like companies was was more than like
  392. 12:00any of us could handle and we did need
  393. 12:02to hire a real go to market team. I
  394. 12:04think at that point we realized hey
  395. 12:05we're actually solving a real pain for
  396. 12:07the market and we are providing a lot of
  397. 12:09value. So that was that happened very
  398. 12:11quickly like within months of the
  399. 12:12product coming out we had a lot of
  400. 12:14companies reaching out to how could they
  401. 12:16run the product uh for themselves and
  402. 12:17and use it for their large code bases.
  403. 12:20Yeah, I think the word product market
  404. 12:21fit is probably something I don't like a
  405. 12:23lot because it it creates a lot of like
  406. 12:25overconfidence. I think what what you
  407. 12:27have as product market fit in one day,
  408. 12:28if you don't continue to innovate and
  409. 12:30you have competitors, is very quickly
  410. 12:31going to become a commodity where you
  411. 12:33don't have product market fit. So, you
  412. 12:34need to be very paranoid. I always like
  413. 12:36to tell the company we're probably going
  414. 12:38to fail. And I think the good thing
  415. 12:39about telling the company we're probably
  416. 12:41going to fail is it creates enough sort
  417. 12:43of energy in the company to find the
  418. 12:46next thing that keeps you as a
  419. 12:48differentiated product and and survive
  420. 12:50in a in a space. But I think in general
  421. 12:52like product like I feel like if you
  422. 12:54have pull from the market, it's very
  423. 12:55hard to fake it. It's it's very obvious
  424. 12:57when you have it and very obvious when
  425. 12:59you don't. And I'll tell you this like
  426. 13:01even when we were making a couple
  427. 13:02million in revenue with the GPU
  428. 13:04virtualization business is very obvious
  429. 13:05we didn't have product market fit. It's
  430. 13:07not a scalable company. you don't really
  431. 13:08understand how you can grow revenue and
  432. 13:10find find a way to like provide more
  433. 13:12value at scale. So very quickly I think
  434. 13:15a lot of large companies were starting
  435. 13:16to reach out to us for a variety of
  436. 13:18reasons for security reasons because we
  437. 13:20were able to run models ourselves. We
  438. 13:21were able to run it in a secure way. The
  439. 13:23other thing is they wanted the systems
  440. 13:25to work with their complicated sort of
  441. 13:27code bases. Right? Their code bases are
  442. 13:29not simple. Some of our customers have
  443. 13:30tens of millions of lines of code in a
  444. 13:32single codebase. and we started building
  445. 13:34more and more technology to make it
  446. 13:35possible that we would give very
  447. 13:37personalized suggestions to them
  448. 13:38regardless of where they stored their
  449. 13:40code. I guess very quickly we started to
  450. 13:42get to 100 customers. It went from going
  451. 13:44from zero to 100 customers probably
  452. 13:46within months um at the company. I think
  453. 13:48the feedback that was kind of
  454. 13:49interesting to us is how important just
  455. 13:51small details were right to the entire
  456. 13:53user experience. We started off as an
  457. 13:55infrastructure company and we needed to
  458. 13:56kind of become a product company which
  459. 13:58is a big kind of change. You know basic
  460. 14:00things like the latency of the product
  461. 14:01how quickly we provide suggestions was
  462. 14:03quite important. You know just basic
  463. 14:04things if we do too much computation on
  464. 14:06the user's machine and it kind of like
  465. 14:08makes the machine too slow that's a
  466. 14:10unacceptable user experience that is not
  467. 14:12something we would think about if we
  468. 14:13were purely a server-based application.
  469. 14:14So there were kind of new things as a
  470. 14:16company we needed to learn because the
  471. 14:18space we were operating in was
  472. 14:19fundamentally different. I think
  473. 14:21basically what happens is once a company
  474. 14:23gets bigger it's possible that you start
  475. 14:25losing the reason why you had a good
  476. 14:27product to start with right you start
  477. 14:29listening actually in some ways too much
  478. 14:31customer obsession might be bad in that
  479. 14:33if you do things that all of your
  480. 14:35customers ask you to do sometimes you
  481. 14:37might like incrementally iterate to
  482. 14:39actually build a really bad product and
  483. 14:41it's actually better off to completely
  484. 14:42change the paradigm in a way that your
  485. 14:44customers don't really ask for because
  486. 14:46you you suspect it could be helpful. So
  487. 14:48there's like a little bit of a healthy
  488. 14:49amount of don't just listen to random
  489. 14:51people about what you should build.
  490. 14:53Listen to your customers, but don't
  491. 14:54listen to your customers in how you
  492. 14:56should build it. There could be a way
  493. 14:57you could build it in a way that is very
  494. 14:59transformational, but not something that
  495. 15:00they would even expect. One of the
  496. 15:02things about our product is we have a
  497. 15:03pretty large individual product and we
  498. 15:05have ways in which our users can
  499. 15:07communicate with us, you know, either
  500. 15:08online on social media. Uh, and I guess
  501. 15:11I I look through it fairly frequently to
  502. 15:14understand what the pain points our
  503. 15:15users have about the product. Also at
  504. 15:17the same time I guess like everyone at
  505. 15:18the company uses the product almost
  506. 15:20day-to-day which is maybe a unique
  507. 15:21aspect of the product right there are
  508. 15:23probably a lot of AI tools out there but
  509. 15:25the people building the tools are not
  510. 15:27using them all the time whereas the
  511. 15:28people that build our products literally
  512. 15:30use our product to build the product
  513. 15:31whether it be windsurf everyone builds
  514. 15:34software on windsurf for windsurf which
  515. 15:36is a unique way for us to actually get
  516. 15:38feedback a lot of our feedback is
  517. 15:40internal and if no one at the company
  518. 15:41likes a part of the product it's
  519. 15:43unlikely that a lot of people outside of
  520. 15:45the company are going to like the

  521. Running 200 People Like 10 - Windsurf's Operating Philosophy

  522. 15:50No, I I always try to run the company as
  523. 15:53like the smallest company it can be. But
  524. 15:55I think the goal of a company is not to
  525. 15:56be a small company. I think the goal of
  526. 15:58a company is to actually provide the
  527. 16:00most value in a space. If the right way
  528. 16:02to do that is be a bigger company. I
  529. 16:04think that's like the right thing you
  530. 16:05should sort of do. One of the operating
  531. 16:06principles we sort of have at the
  532. 16:08company is we try to run the company
  533. 16:09fairly lean. And that doesn't mean we
  534. 16:11have very few people. It just means that
  535. 16:13for a given amount of ambition, we are
  536. 16:15the smallest company we could possibly
  537. 16:16be. And that means that people are
  538. 16:18largely speaking underwater most of the
  539. 16:20time. Hopefully like we don't have large
  540. 16:22amounts of people that have too little
  541. 16:24to do, right? The goal is everyone
  542. 16:26should have too much to do. And because
  543. 16:27of that, that forces rapid
  544. 16:29prioritization inside the company with a
  545. 16:31lot of urgency that triggers whether or
  546. 16:33not we should hire people. And honestly,
  547. 16:35that itself is a good forcing function
  548. 16:36of how you should scale a company. You
  549. 16:38scale a company when everyone is
  550. 16:40underwater and the moment people are no
  551. 16:42longer underwater, we don't scale the
  552. 16:44company anymore. I think in in terms of
  553. 16:46just some of the challenges obviously
  554. 16:48like as the company gets bigger,
  555. 16:49communication gets harder. So there are
  556. 16:51some processes in place in terms of you
  557. 16:53want to ship a feature, it's no longer
  558. 16:55just one person did everything end to
  559. 16:57end. There are like many different
  560. 16:58parties involved. Uh but I think we
  561. 17:01added more process in place to make sure
  562. 17:02that even that could happen much faster.
  563. 17:04If you run the company in a way where
  564. 17:06everyone has their own set of
  565. 17:07priorities, I think it's it's cool.
  566. 17:09Everyone's kind of having a good time.
  567. 17:11But I think, you know, this boils down
  568. 17:12to why do companies work? I think they
  569. 17:14work if you do one or two things really
  570. 17:16well. Maybe even one. Forget about two,
  571. 17:18maybe even one. The problem is then
  572. 17:20actually if you were to imagine in a lot
  573. 17:23of cases, it's very hard to just have
  574. 17:25infinitely many people doing one thing
  575. 17:27really well. So that naturally reduces
  576. 17:28the number of people that you have at
  577. 17:30the company. Yeah. I guess I guess the
  578. 17:32the issue is naturally when you have
  579. 17:34more people, they're going to be more
  580. 17:35priorities, right? Like people are not
  581. 17:37malicious, but they want to have
  582. 17:39something to do. If there are enough
  583. 17:40people to do what the original intent of
  584. 17:42the company was, they're going to find
  585. 17:44other things to do and that's going to
  586. 17:45cause problems internally as well. One
  587. 17:47of the things that I've sort of done is
  588. 17:48I still interview everyone that joins
  589. 17:50the company, make sure that they go
  590. 17:52through a culture fit. I think that's
  591. 17:53actually like quite important to me to
  592. 17:55make sure that we are adding people that
  593. 17:57are going to like maintain and preserve
  594. 18:00and maybe even even just embrace the
  595. 18:02culture we have inside the company. So I
  596. 18:04think that's like honestly honestly like
  597. 18:05a very key thing that I've not given up
  598. 18:07even as the company has scaled you know
  599. 18:08close to 200 people. I I think actually
  600. 18:11it's the same principles that made the
  601. 18:12company work when it's small. I I think
  602. 18:14big companies are not that different
  603. 18:15than small companies. Maybe it's hard to
  604. 18:16operate like a small company when you're
  605. 18:18a big company but I think if you were to
  606. 18:19ask every big company they want to
  607. 18:20operate like a startup. I think the very
  608. 18:22hard thing is once again intellectual
  609. 18:23honesty. It's very hard as the company
  610. 18:25gets bigger for for a large group of
  611. 18:27people to analyze what they're doing and
  612. 18:29to say, "Hey, we should stop working on
  613. 18:31this or should work on something else."
  614. 18:32It gets very hard because people feel
  615. 18:34some sort of safety as the company gets
  616. 18:36bigger. But the reality is like no
  617. 18:38company is really safe regardless of the
  618. 18:40size. And that's how every company
  619. 18:41should be operating. They should be
  620. 18:42operating as if almost existential
  621. 18:44dread. Not in that it's causing
  622. 18:46paralysis, but in that it causes the
  623. 18:47agency to kind of figure out what the
  624. 18:49next thing to do is. Yeah. I think
  625. 18:51people get they feel safety,
  626. 18:53psychological safety by feeling that
  627. 18:56what they're working on is the most
  628. 18:57important thing, right? I think, you
  629. 18:59know, people it's very hard to keep in
  630. 19:00your head, hey, like let me go work hard
  631. 19:02on something, but all the while it may
  632. 19:03not be that useful for the company. And
  633. 19:04I think as a company gets bigger, it's
  634. 19:06very hard to have people that just still
  635. 19:08don't have the psychological safety.
  636. 19:10Maybe one interesting thing about the
  637. 19:11company is I think companies have like
  638. 19:13two sides attached to them, which is
  639. 19:15that you need some form of irrational
  640. 19:16optimism. The reason why you need
  641. 19:18irrational optimism is without some
  642. 19:20amount of optimism, there's no reason a
  643. 19:22startup ever wins at anything. A big
  644. 19:24company has more capital, more
  645. 19:25resources, more distribution. So you
  646. 19:27need to believe you can do something
  647. 19:28better, right? With great people, uh you
  648. 19:30can somehow build something that's
  649. 19:32generational. You have to believe that.
  650. 19:34Uh which is in most cases not true,
  651. 19:36right? So you somehow need to believe
  652. 19:37that. But also at the same time,
  653. 19:38sometimes that's just not going to be
  654. 19:39the case. And you need to be very
  655. 19:41realistic. And I think this comes with,
  656. 19:42you know, uncompromising realism. Like
  657. 19:44you need a healthy amount of both. You
  658. 19:46need you need the optimism to go out and
  659. 19:48do something when everyone else believes
  660. 19:50it's not that valuable because if
  661. 19:51everyone else believed it was valuable,
  662. 19:53the big company would do it and you
  663. 19:54would still lose. But all the while you
  664. 19:56also need to agree most ideas are bad
  665. 19:58ideas and you need to kill your ideas
  666. 19:59fairly quickly as well. This is this is
  667. 20:01like this tension that a company is
  668. 20:02always going through. It progressively
  669. 20:04gets harder and harder. The company gets
  670. 20:05bigger. Uh but I think it's a strength
  671. 20:06if a company can do that regardless of
  672. 20:08what size it is.
  673. 20:10Yeah. I think the only thing is that I
  674. 20:13would sort of like to say is build for
  675. 20:14where you think the technology is going,
  676. 20:16not for today, right? And don't build
  677. 20:18probably technologies in places that
  678. 20:20like are as good as they will be given
  679. 20:22where the technology is today. It feels
  680. 20:24like it's going to be very hard to
  681. 20:25differentiate in the long term. And
  682. 20:26that's a hard thing to do, right? Uh
  683. 20:28that means you you you actually need to
  684. 20:30be willing to work on something that
  685. 20:31doesn't feel like it's working for
  686. 20:32maybe, you know, an indeterminate period
  687. 20:34of time. I think what this means is
  688. 20:35these kind of short-term heristics that
  689. 20:37you invest deeply in that make a model
  690. 20:39today work are probably actually have
  691. 20:41massive diminishing returns because the
  692. 20:43next model is going to be much better
  693. 20:44and the heristics are going to be
  694. 20:45unnecessary. I think the things that are
  695. 20:47important are how do you actually take
  696. 20:50advantage of the fact that you do have a
  697. 20:51lot of users and build better
  698. 20:52experiences for them and actually
  699. 20:54because they are using your product
  700. 20:55you're able to build better and better
  701. 20:57experiences for them and these
  702. 20:58experiences are better and learned in a
  703. 21:01way they're actually learned from the
  704. 21:02way people use the product and I think
  705. 21:04you should be focusing on that not on
  706. 21:05kind of cosmetic things that make a
  707. 21:08model kind of magically do something
  708. 21:10that probably is going to happen anyways
  709. 21:12in 6 months.
How I Built a $1.3B Startup by Pivoting Fast | Windsurf, Varun Mohan, Co-Founder & CEO — Transcriptly