Transcriptly

Vibe Coding All Day? Stop Delivering PrototypesㅣGraphite, Merrill Lutsky

EO09:43Added Aug 31, 2026

Merrill Lutsky is the co-founder and CEO of Graphite, an AI code review company. Graphite has raised $81 million in funding and is recognized as a rapidly growing enterprise, having achieved 20X growth last year.

Watch on YouTube →
Contributed by 刘嘉琪

Transcript

Transcript format

  1. Intro

  2. 00:03that's really been exciting about
  3. 00:05building this company, especially over
  4. 00:07the past year as code generation has
  5. 00:08gotten so good. As agents get better and
  6. 00:11better at generating code, the the way
  7. 00:13that you're interacting with them
  8. 00:14probably isn't going to look like how we
  9. 00:16we write code today. it's v 0ero bolt
  10. 00:19new replet agent like something like
  11. 00:21that where you know where you're
  12. 00:23interacting with these agents and
  13. 00:24prompting them and and iterating on
  14. 00:26their experiences and it's almost more
  15. 00:29analogous to code review than it is to
  16. 00:31how we would write code previously and
  17. 00:34that's why we think like owning code
  18. 00:36review as as the starting point is what
  19. 00:38matters you know far more so than the
  20. 00:40fist fight that's happening over like in
  21. 00:42loop code generation right
  22. 00:48Hi, I'm Mel Lutzky, co-founder and CEO
  23. 00:50of Graphite. Graphite is the code review
  24. 00:52platform for the age of AI. We take
  25. 00:55best-in-class workflows and tooling from
  26. 00:57companies like Google and Meta and
  27. 00:58combine them with our agentic AI code
  28. 01:01review companion to help engineers uh
  29. 01:03create, review, and merge code changes
  30. 01:05faster than ever. We work closely with
  31. 01:07the team at Enthropic um to power our AI
  32. 01:10code review agent um called Diamond. The
  33. 01:13series B that we just announced was $52
  34. 01:16million. Um, and Graphite has grown 20x
  35. 01:18in the past
  36. 01:20[Music]

  37. What If Everyone Could Use Big Tech’s Internal Tools?

  38. 01:24year. Graphite started on a completely
  39. 01:28different idea. We actually we actually
  40. 01:30had sort of two one small pivot in the
  41. 01:32beginning in the first three months and
  42. 01:33then then the iOS tool. We'd hired in
  43. 01:36the process then a few engineers uh a
  44. 01:39few of Tomas's ex-coorkers from Meta and
  45. 01:42this we'd had this really peculiar thing
  46. 01:44that happened when when they onboarded.
  47. 01:46So they'd start on their first day we'd
  48. 01:48get them set up on GitHub. We'd show
  49. 01:50them how we do code review at the
  50. 01:52company and all of them had this kind of
  51. 01:55allergic reaction to it immediately and
  52. 01:57they all said you know this feels like
  53. 01:59so much slower than the tooling that I'd
  54. 02:00had at Meta. I I feel like a caveman
  55. 02:03using this. Google and Meta have both
  56. 02:06built amazing internal tools to help
  57. 02:08their their teams of thousands or tens
  58. 02:10of thousands of engineers um all working
  59. 02:12on on one monor repo at once even across
  60. 02:16across time zones and across you know
  61. 02:18spanning these massive code bases. And
  62. 02:20um if you go outside of those walls
  63. 02:22though u it's very hard to to get that
  64. 02:25same level of of productivity and uh
  65. 02:28have that same that same tool chain. And
  66. 02:30that was a lot of why, you know, why we
  67. 02:32started Graphite was we saw kind of the
  68. 02:34the gap between what we'd had at at the
  69. 02:36large companies that that all us and the
  70. 02:38founding team had worked on and what was
  71. 02:40available to every company when they're
  72. 02:43just getting started. And um that was
  73. 02:44really what inspired us to to build
  74. 02:46graphite. I think it's so important to

  75. The Most Important Factor for Success

  76. 02:49like deeply understand and care about
  77. 02:51the user that you're building for.
  78. 02:53You're going to spend if your company is
  79. 02:55successful, you know, 10 plus years
  80. 02:57talking to these people every single
  81. 02:58day. And if you don't genuinely care
  82. 03:01about making their lives better, about
  83. 03:02solving their problems and about
  84. 03:04building uh product experience they
  85. 03:07really enjoy using and get a ton of
  86. 03:09value from, it's going to be really hard
  87. 03:11to just sustain your your level of
  88. 03:12enthusiasm about that. And one of the
  89. 03:14the things that's so fun about graphite
  90. 03:17and one of the best parts about getting
  91. 03:18to work on this is that it is something
  92. 03:20that we all on the team use every day. A
  93. 03:23lot of like great companies use every
  94. 03:24day. our friends at at these companies
  95. 03:26are are using and are constantly texting
  96. 03:28me with feedback. It's just fun to be
  97. 03:30able to to build something that that's
  98. 03:32for um a user that we're so you know
  99. 03:35that we're so deeply familiar with and
  100. 03:37um that we genuinely care about and and
  101. 03:39want to like make their lives better. So

  102. How Is Graphite Different?

  103. 03:41in the beginning we started with tool
  104. 03:44for creating stacked pull requests. The
  105. 03:46idea of stack pull requests um is that
  106. 03:48instead you you break and having that
  107. 03:50all be one pull request um instead you
  108. 03:53break that up into a series of small
  109. 03:56dependent changes um that all sort of
  110. 03:58chain together and build upon each
  111. 04:00other. And the advantage of this is that
  112. 04:01you know first of all nobody likes to
  113. 04:03review like a thousand line pull
  114. 04:05request. that's like, you know, there's
  115. 04:06that meme that's like, you know,
  116. 04:07thousand lines like looks good to me.
  117. 04:10And I think that that really is is what
  118. 04:11you see play play out in the data. Like
  119. 04:14nobody wants to to like give a detailed
  120. 04:16review to a change that big. Um that's
  121. 04:18really why, you know, why stacking is is
  122. 04:21so powerful is it just it gives you so
  123. 04:23much time back um that you would have
  124. 04:25spent kind of sitting on your hands or
  125. 04:26trying to find something else to build.
  126. 04:28And we've seen like really meaningful
  127. 04:30improvements in in development speed at
  128. 04:32many of our top customers. ASA was the
  129. 04:35one where where we save 7 hours per
  130. 04:37developer per week. At RAMP, we
  131. 04:39increased their development speed by 3x.
  132. 04:42One of our adviserss kind of famously
  133. 04:43said like once you stack, you don't go
  134. 04:45back. And I think that that's really
  135. 04:47true in terms of the both like the user
  136. 04:49excitement and then also just the the
  137. 04:51retention numbers and and you know the
  138. 04:53engagement that we see with graphite
  139. 04:55pretty much every

  140. Why Code Generation Isn’t Enough in the Gen AI Era

  141. 05:02day as every engineer has tools like you
  142. 05:05know cursor and windsurf and copilot and
  143. 05:08even some of these more agentic code
  144. 05:09generation tools now uh they're able to
  145. 05:12like write code faster than ever uh they
  146. 05:14can they can do or they're creating more
  147. 05:17code changes than than they ever have
  148. 05:18before. That's only really the first
  149. 05:20part of the story. I think even what
  150. 05:22we're seeing with with vibe coding right
  151. 05:24now is a lot of those products are kind
  152. 05:26of MVP like there. You still need to
  153. 05:29take that change and get it through code
  154. 05:31review. You still need to test it. You
  155. 05:32still need to deploy it and measure the
  156. 05:34impact. and that whole kind of outer
  157. 05:36loop process um where you're
  158. 05:38collaborating with your teammates that's
  159. 05:40like a huge part of of your life you
  160. 05:42know almost as much in many cases as the
  161. 05:45inner loop of of how I'm like writing
  162. 05:47code testing locally and getting it
  163. 05:49ready to uh to like push up to to a pull
  164. 05:52request. So I think that as you know as
  165. 05:55you have like this this idea of like
  166. 05:57vibe coding and AI code generation it's

  167. What Should Engineers Do in the Gen AI Era?

  168. 06:00almost not enough to be just you know
  169. 06:01engineer anymore like the best engineers
  170. 06:04are also great product thinkers and
  171. 06:06great business thinkers great designers
  172. 06:08in some cases it's really it used to be
  173. 06:10the case I think that the limiting
  174. 06:12factor on creating anything was was you
  175. 06:15know asking like how do I actually build
  176. 06:17this and how do I implement it now you
  177. 06:20have these great tools that can help you
  178. 06:21answer that question and kind of get to
  179. 06:23an early version like a working
  180. 06:26prototype of of something like quite
  181. 06:28quickly. So then the question shifts
  182. 06:30from being like how do I build this to a
  183. 06:34little bit further upstream of like what
  184. 06:36should I build in the first place that's
  185. 06:38actually going to be of value? you know
  186. 06:40how do I actually interact with this and
  187. 06:42what is the optimal experience here and
  188. 06:44then how do I differentiate that and how
  189. 06:47do I in a world where like anybody can
  190. 06:49you know go on replet agent and prompt
  191. 06:51you with a few prompts make you make
  192. 06:53make like a working prototype of
  193. 06:55something how is that different from
  194. 06:57anybody else's like vibecoded version of
  195. 06:59this and I think that that software
  196. 07:02engineering looks a lot more like you
  197. 07:03know reviewing or even designing or or
  198. 07:06like you pming in the future versus like
  199. 07:09writing code as we know it today.

  200. Building the Right Thing at the Right Time

  201. 07:17the transformation of software
  202. 07:18engineering uh with AI code generation
  203. 07:21is has just been the sort of seismic
  204. 07:23shift that's happened in you know as
  205. 07:25we've been building the company and you
  206. 07:27know we were fortunate in that we were
  207. 07:28working on solving a problem that was
  208. 07:31now made 10 times worse by AI code
  209. 07:33generation code review has always been
  210. 07:36and this outer loop of review and
  211. 07:38testing and deployments has always been
  212. 07:40a bottleneck at large companies it's
  213. 07:41always been an area that that needs
  214. 07:43better tools And that was true before
  215. 07:46AI, but now with with all these great AI
  216. 07:49code generation tools, that need is is
  217. 07:52all the more acute. And I think that's
  218. 07:53that's one of the things that's been I
  219. 07:55think we were lucky that we were
  220. 07:56building kind of the the right thing at
  221. 07:58the right time. U because we built uh we
  222. 08:01built tooling for accelerating the outer
  223. 08:03loop and we got that into a lot of great
  224. 08:06enterprise customers. We also built the
  225. 08:08trust with those organizations to be
  226. 08:11able to now deploy AI to help them
  227. 08:13review code faster. Uh I think it's
  228. 08:15really hard if you're just starting an
  229. 08:17AI code review company to convince an
  230. 08:20enterprise or convince a company like
  231. 08:21Shopify to give you access to their
  232. 08:23entire GitHub repo from day one. It's
  233. 08:25taken us a long time to build up that
  234. 08:27level of trust. And I I think there are
  235. 08:29a few factors there. One is just like
  236. 08:31building just adding value progressively
  237. 08:33over time, shipping features constantly,
  238. 08:36like showing the customers that we work
  239. 08:38with early on that we are consistently
  240. 08:40able to like deliver on our promises and
  241. 08:42you know continue to just add more and
  242. 08:44more great features to the graphite
  243. 08:45platform. Um we spend a lot of time
  244. 08:47making sure that that everything we do
  245. 08:49is like you know best practices for
  246. 08:51security. We're pentested very
  247. 08:53regularly. Like we really want to invest
  248. 08:55in in making sure that that that piece
  249. 08:57is is central to what we're doing you
  250. 08:59know. And then and then part of it is
  251. 09:00just climbing the ladder. I think you
  252. 09:02have to you have to start with like
  253. 09:04smaller companies first like you're
  254. 09:06always going to have kind of more you
  255. 09:07know smaller companies that are willing
  256. 09:09to adopt tools tools in the beginning
  257. 09:11and then like building up trust over
  258. 09:13time and I think now that snowball has
  259. 09:15really like started to roll and um we
  260. 09:17have you know some of some of the
  261. 09:19largest technology companies out there
  262. 09:21that are are trusting graphite every day
  263. 09:22and u you know we're we're really
  264. 09:24privileged to be able to serve
Vibe Coding All Day? Stop Delivering PrototypesㅣGraphite, Merrill Lutsky — Transcriptly