# Graphite’s Merrill Lutsky on Revolutionizing Code Review for the AI Era **Guest:** Merrill Lutsky **Role:** CEO and Co-founder, Graphite **Host:** Christine Esserman **Season:** Season 3 **Episode:** 8 > In this episode of Spotlight On, Accel’s Christine Esserman sits down with Merrill Lutsky, co-founder and CEO of Graphite, to discuss how his company is redefining code review for the AI era. Merrill shares how he and his co-founders have fostered an unconventional yet deeply effective approach to building trust through weekly Monday dinners and co-founder “therapy” sessions. They discuss the unique challenges of building a developer tools platform inspired by Meta’s engineering culture and how Graphite’s AI-powered code review companion, Diamond, enables teams to work smarter and faster. From navigating pivots to scaling a fast-growing AI startup headquartered in New York City, Merrill offers insights into the messy, exhilarating process of building a company that’s reshaping how developers collaborate in the age of AI. --- ## About this episode In this episode of Spotlight On, Accel’s Christine Esserman sits down with Merrill Lutsky, co-founder and CEO of Graphite, to discuss how his company is redefining code review for the AI era. Merrill shares how he and his co-founders have fostered an unconventional yet deeply effective approach to building trust through weekly Monday dinners and co-founder “therapy” sessions. They discuss the unique challenges of building a developer tools platform inspired by Meta’s engineering culture and how Graphite’s AI-powered code review companion, Diamond, enables teams to work smarter and faster. From navigating pivots to scaling a fast-growing AI startup headquartered in New York City, Merrill offers insights into the messy, exhilarating process of building a company that’s reshaping how developers collaborate in the age of AI. ‍ **Conversation Highlights** - [00:32] What is Graphite? Merrill explains the AI-powered code review platform transforming developer workflows. - [01:02] Graphite’s $52M Series B and its impact on scaling operations. - [03:09] Merrill’s first startup experience and how it shaped his approach to founding companies. - [04:27] The framework Merrill uses to evaluate team, idea, and market timing before launching new ventures. - [05:33] Building trust and collaboration with co-founders—how Monday night dinners and co-founder therapy strengthened their dynamic. - [11:17] Pivoting to Graphite: The unexpected demand for better code review tools inspired by Meta’s engineering culture. - [16:17] The origin of the name “Graphite” and its connection to iterative workflows. - [27:17] Navigating seismic shifts in software development with AI—how Graphite adapted to the rise of LLMs. - [32:15] Launching Diamond, Graphite’s AI-powered code reviewer, and how it passed the “Dave Test.” - [42:40] Merrill’s personal pursuits—running marathons and DJing in New York—and how they fuel his energy as a founder. ## Transcript Merrill (00:00): We really want to build the best platform where that activity takes place and kind of be the home screen that developers are living in pretty much all of their professional lives. Christine (00:10): Welcome to Spotlight on a podcast about how companies are built from the people doing the building one messy, exhilarating decision at a time. Christine (00:19): Welcome to Spotlight On. I'm Christine Esserman, and I'm so excited to be here with Merrill Lutsky, the CEO of Graphite. Merrill (00:25): Thanks for having me, Christine. Christine (00:26): Of course. Merrill, maybe just to start off, can you tell me a little bit more about what Graphite is for those that dunno? Merrill (00:32): Yeah, happy to. So Graphite is a code review platform for the age of ai. We take a combination of best in class workflows and tooling inspired by what the developer platforms of companies like Google and Meta, and combine that with an agentic AI code review companion that gives developers feedback in seconds on every single code change that they're putting up, helping teams merge and review code faster than ever. Christine (00:57): Amazing, and congratulations because I know you just announced your $52 million series B Merrill (01:02): Thank you Christine (01:02): Led by Excel and we feel so fortunate to be working with you and the entire graphite team. So big week for you, huh? Merrill (01:08): Thank you. Thank you. Yeah, it's been an incredible week and it's been so fun to see the response of so many teams to what we're doing with Iman now. Christine (01:14): Amazing, amazing. Well, before we dive into some of the new product updates, let's rewind to your background if you're okay with it. This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 2 of 2 Christine (01:22): Yeah, Christine (01:23): You have quite the past and one of my favorite things about you is just how well-rounded you are. You're a founder, you're incredibly dedicated to the company, but you're also an incredible athlete and musician and you keep yourself very busy. So I'd love to go into your personal background maybe just to start, this isn't your first company. Walk us through your first company, how you initially got the startup bug and why you want to do it again. Merrill (01:49): Back in 2013, my first startup was a company that I started with. One of my classmates at Harvard, Eric Schloss, who's actually funny enough is now one of our main points of contact on the ENG team over at Philanthropics. So it's kind of fun to be working with him closely. Again. Back then we started a customer feedback tool for hotels and restaurants. This kind of grew out of a project we'd built for our dining halls to rate the food and the Harvard dining halls and had expanded to local businesses in Cambridge and then we'd gotten into yc, took a year off of school and moved out here to come and build that company and really got a whirlwind education in how to build a company, how to hire a team, how to find customers. We were going door to door to all the hotels in San Francisco and talking to the managers and getting kicked out of some of them and really got the startup building experience from that. And although that company wasn't really a huge success, we had a small exit out of that. I think it taught me a lot about building a team, building a product, what kind of markets that I wanted to build in and really left me excited about both building great technology and building companies again. Christine (03:09): That's amazing. Well, Harvard Dropout, what did your parents say when you decided you were moving across the country to build something at such a young age? Merrill (03:17): Well, my parents are both professors at a small liberal arts college, so I've always gotten the question of when are you going to go do your PhD? And then at some point that became, oh, maybe an MBA is going to be the right thing. But no, actually they were really supportive. I think YCU was actually a very helpful, very helpful tool for that and that I think it de-risked the experience of dropping out and gave really a good excuse to really focus all of our energy and come and learn how to build great companies. Was really great to work with. Our group partners, Jeff Ralston, Kevin Hale there again, learned a ton. I kind of call that my study abroad year in Silicon Valley and then fortunately ended up making the decision to go back and finish my degree unlike some of my more successful dropout compatriots. And that was actually what led me to meeting my now co-founders, Tomas and Greg. Christine (04:12): Wow, amazing. So you go back to Harvard, you've learned all these startup lessons, you exited a company and you still have energy to do it again. At that time, what were you looking for? What was the biggest learning on what was going to be different the second time around? Merrill (04:27): This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 3 of 2 Yeah, I think I developed a framework at some point around team idea and market timing and having to have at least two of three and ideally three of three on those before I was to start something again. And it took a while before I really felt like I had that all three of those kind of all materializing at once and I was trying to be very intentional about not forcing it. I think it's very easy to sit down and try to just generate startup ideas or go founder dating and try to find somebody to start a company with. And I think that the first experience in particular just really taught me the importance of having it evolve naturally and come from a problem that you care about, a user that you deeply understand, and most importantly a team of founders that you really care about, you have deep trust with that you're excited to work with and that you can see yourself building with for 10 plus years. Christine (05:24): Yeah. Tell me more just about your team, your co-founders, how you compliment each other, how you challenge each other, and just the pros and cons of having a three person fatty team. Merrill (05:33): Yeah, yeah, we talk about it as the tripod, so you have a very strong support system of even if one of us has a personal thing or family thing or even just isn't feeling a hundred percent on a given day, I think it's super helpful to have not just one but and two other co-founders to support you on that. And yeah, I think the three of us really, really compliment each other. Well, going back to how we met, Tomas and Greg were project partners all through the hard CS courses at Harvard. They talk about, one of them would be working all day on their operating systems and then they'd hand off and actually their interest in code review came from experiences that they had in that class where basically they'd spend hours and hours trying to debug something and one of them would push changes and overwrite the others and they just had, at some point, I think Tomas told Greg, okay, we're going to start doing code review now because we're just making too many silly mistakes Christine (06:32): That sounds like something from the top and deciding what to do. Merrill (06:37): So I think that really that plus then I met the two of them coming back into the startup community on campus and spending a lot of time thinking about side projects. Tomas and I used to work from each other's offices on the weekends in New York all the time when I had first graduated, really kind of building a common language around what companies we wanted to build and how we thought about different problems in tech. And I also think one of what was, I guess a challenge at first, but also one of the great strengths of us as a founding team is we all come from different engineering and different company cultures. So I really cut my teeth at Square where I went after the first company that I built. I worked on the team there that built the Square feedback system. So if you see this is still one of the most used products I've ever worked on, if you get the happy and sad face on the square receipt, that was kind of in some ways the legacy of the first startup that I built. (07:38): And I think being at Square really taught me a care for craft and quality of product and really focusing on creating a brilliant user experience and also the power of applying great design and product thinking to an area that's traditionally underserved. I think before Square and Stripe and a few others, payments was just an incredibly unsexy space. No designer wanted to work on a payments product. And now for This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 4 of 2 some time in particular, it was one of the hottest spaces for design. And I think we're now seeing that trend play out with dev tools where there's now just beginning to be a whole new guard of companies like Linear and CEL and Planet Scale and a few others that are really paying deep attention to design in dev tools and crafting experiences that are orders of magnitude better than what's existed before. So I think that was a lot of the DNA that I brought. (08:31): Tomas really brought from spending a lot of time at Meta. He really saw the coder view platforms that Meta had built and also just the value of shipping quickly, iterating really moving fast and with urgency that meta engineering culture has. And then Greg, I think at Airbnb saw a much different way of building a company under Brian, and again, similar desire for great design and product thinking, but much more methodical on the infrastructure side and making sure that you're providing things that people are relying on every single day and that they can trust. And I think the three of us coming from those three different perspectives and figuring out what elements of each culture we wanted to bring with and which ones we wanted to leave behind, I think has been really, really helpful for us in forming what's now become kind of the graphite culture Christine (09:25): On just your co-founder dynamic. How have you been able to maintain your relationship through different times? I think one thing you do that's really cool is the Monday night dinners together, but I'd love to hear more strategies on just maintaining a healthy relationship with your co-founders as the company scales. Merrill (09:42): Yeah, I think that that's so important and something that's really helped us to get through some of the hardest moments of the company in particular. So early on we have a regular dinner that we do as founders. I think that really creates space for us to talk about what's on our minds, how we're thinking about things at a high level, and also just to maintain a great friendship. I think it's hard sometimes when you have such close friends and then you start working together. It changes the nature of the relationship and you really have to invest in keeping that basis of trust and friendship. Another interesting thing that we did early on, we actually found a therapist and all three of us would go to, we'd go to You Christine (10:25): Did couples therapy. Yeah, Merrill (10:26): We did co-founder couples therapy and it was actually early on in terms of learning. I think one of the best things that we did in terms of learning to trust each other, learning how to work together, managing a lot of the tensions and disagreements that came up early in the company's history. And I think that at some point we actually got to the point where we'd have these meetings and there's like, there's not really much to talk about today. It got to a point where we felt like everything was working really well and that's continued to be the case, but I think that was one of the best hacks that we did early on to create that space every week and just force the hard conversations and make sure that we could maintain a good relationship as a founding team. This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 5 of 2 Christine (11:07): I love that you did that. Well, let's transition to the early days of graphite. You have this incredible founding team, you're all technical, you have a ton of skills, how do you decide what to Merrill (11:17): Build? So from the beginning we had the thesis that the biggest tech companies and many that we'd seen and worked at have figured out how to do developer tooling and how to ship product just much faster than what's available to the rest of the world. Actually the first thing we built was more of a loom for developers. So the idea, a lot of companies have built a bug capture tool internally where you can record console logs and network requests and create really rich information for someone who's trying to debug a front-end issue. That was the first thing that we built in the first three months of the company and we built that. We got that in the hands of a few friendly companies that we knew and we just weren't seeing the kind of really regular daily usage or level of excitement about that product as we wanted to see. (12:17): So we pretty quickly made the decision to shut that down. And the second thing, I guess one thing that we learned in doing that was we'd had a lot of asks for help around mobile, and that was particularly mobile quality was very much where there was a lot of pain that we were hearing, and the next idea that we built was inspired by tools that Meta and Uber and a few others had built internally, which is an iOS rollback platform. So something that lets you roll back and native iOS app deployment or release in the same way that you can roll back a bad web deployment in a few minutes instead of having to wait to go through the app store resubmission process and potentially days where you have users with a buggy product in their hands. Just being able to roll back on the end user's device almost instantly is something that they can be really powerful. (13:07): And we were going deep on decompiling and recompiling apple binaries and inserting a feature flag between successive versions and playing around with lots of undocumented APIs that Apple definitely wasn't happy that we were in at the time, and we actually had a few larger customers using it. We had some larger companies that had our tool live that had given us intent to pay, but it just wasn't, again, wasn't the type of excitement that we really wanted to see from an early tool. We'd roll it out, they'd be sort of hesitant to use it if something messed up, it would take a month or two weeks until the next release cycle. It really just wasn't the type of excitement and traction that we wanted to see from it. Graphite really started when we hired our first few engineers. So because the iOS app, the iOS rollback platform was going reasonably well, we started hiring a team. (14:11): We brought in a couple of temas ex-coworkers from Meta and pretty much when they got here, we set them up on GitHub. We teach them how you do code review or how the rest of the world does code review, and they had this universal allergic reaction to it. It was instantly like, how do I do this? This feels so much slower and less efficient than the tooling that I'd had at Meta. One of them famously said, I feel like a caveman using this tool. And it was such a point of frustration for them that they actually ended up packing together a little CLI tool to use for just to approximate this stackpole request workflow that they'd had at Meta and that that was what grew into the first version of graphite. And something really interesting happened where I think at some point they mentioned to a few X-ed coworkers that they built this thing and all of a sudden we started getting inbound from random X meta engineers. (15:14): This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 6 of 2 That was where they would say, Hey, I heard you built this stack pull request tool. Can you give it to me? And sometimes we'd even be on a demo for the iOS platform and there'd be an X meta engineer on the other side who would say, this iOS thing is cool, but tell me about the Stack Diffs thing. Tell me about this tool that you've built. Can I buy that? Can you guys make that for us? And we'd say, no, it's hardcoded for our repo. This is just an internal thing that we've built. We're not trying to commercialize this. And they're like, no, I really want that thing. And that level of excitement built up over the course of a few months, and eventually it just became too loud to ignore and we made the decision early. It was still fairly early on at the company to pivot to focusing entirely on building code review and quickly kind of amassed our first 20, then 30, then 50, then a hundred early users at some great companies with amazing engineering teams. And yeah, then it was kind of off to the races from there. Christine (16:15): Where did the graphite name come from? Merrill (16:17): It came from, there were a couple of things, a couple aspects of it. We really liked kind of the layers of graphite of having stacked prs and layers and sheets of graphite. We also like the idea of working in pencil instead of pen and kind of rapidly iterating on things. We liked hexagons and the.dev domain was available most importantly. Christine (16:39): Amazing. Amazing. Okay, so you have early product market fit. What do you do now? Do you hire more? Do you try to build out a sales team? What's the next step Merrill (16:49): For us? In the beginning, one of the things that was really interesting about Graphite was it was this individual engineering tool. It wasn't really something that we felt was something we could sell immediately to ENG teams. We built it on top of GitHub intentionally so that mostly just because we were using GitHub at the time and wanted to keep using GitHub for hosting, but wanted this workflow for ourselves. So we had this tool, it was very bottom up adopted. The spread pattern that we were seeing was very much individual engineers sharing it with each other, and what we realized was that there was a lot more, if you looked at the meta and Google platforms that inspired it, it's not enough to just create a command line tool for this. There were actually a bunch of pretty good open source tools out there for just creating stack pull requests at the time, and that I think they hadn't really gotten to adoption when we talked to a lot of users of them. (17:53): The reason why they hadn't reached broader adoption was because it's only one part of the story. If you want to work in this way where instead of writing massive prs and waiting for review, you're breaking your changes up into chains of small dependent pull requests and pushing as you go and constantly having a lot more small changes that are flying around at any given time, you need a bunch of other infrastructure as you start to scale that up in order to do it. And if you look at what Meta and Google built there, it really isn't just that first creation part. It is this whole tool chain to support working at that speed. It starts with the CLI and VS code extensions and that local experience, but then it's discoverability of prs. It's organizing the workflow and making sure every engineer knows who are they blocking at any given time, who do they need to ping in order to unblock their own changes? This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 7 of 2 (18:52): It's tools like automations for assigning reviewers and labels and coordinating where this change is in the process. It's emerge queue to effectively merge these stacks of changes all at once and keep your trunk branch green when you have thousands of engineers that are all trying to create changes at the same time. That whole tool chain is really what you need to scale this up. And really the next, we were in close beta for about a year and a half because we wanted to build out the rest of that story and we didn't think that it was truly ready for enterprise scale before we had kind of finished the V one of that tool chain end to end. And that was really where we went next was based on the initial traction and our wait list launch around the initial stack PRS tool. That was when we raised our series A. We started building more of a team and then we really embarked on that journey to build out the rest of the platform. And this was all kind of before LLMs changed everything with coding. Really our focus was just building best in class tooling for helping teams deal with a lot of code changes. And you can see how that, it's been interesting how that foundation has now come into play with all the recent innovations. Christine (20:12): Yeah, it's amazing. And I think that the tool chain is super, super interesting just to hear you talk about it. And I'm just struck by the number of features that are within the graphite core platform and that make up this end-to-end tool chain. But at the time that we're talking right now, you probably have fewer than 15 employees. The team is pretty resource constrained and you're building a pretty broad platform. How do you orient the team just around focus and product sprints and meeting strategically? What's your philosophy on just splitting up the team to make sure that able to focus on all of that? Merrill (20:48): Yeah, it's a really interesting challenge and one of the things that I think is the hardest about what we do is just every prioritization meeting, every roadmap meeting is contentious and it's hard for, we're all just trying to figure out what is the right next thing to build. And I think that it's been a combination of having a clear perspective on what does this end tool chain look like and what does that North Star for what we want to build, how that's changed a lot even in recent months, almost on a weekly basis. I feel like it's changing in terms of what's possible and what we think will be possible soon in terms of how AI comes into play with this, but having a clear north star of that, but then not balancing that with customer feedback and taking into account some of our largest customers, a lot of both our large customers and then also we've had a fantastic amount of feedback from our community since the early days of graphite. (21:49): I think user feedback has been one of the biggest inputs that we've had into our roadmap and really helped us make sure that we're building the right things. And I think you do have to balance those things. If you're too far in just following your vision and not keeping near to the ground of what customers are saying, then you're not going to be building something that they actually want that solves their problems. But at the same time, if you're too far on the just do everything the customers say, then you end up with this really bloated and convoluted product that's not coherent and that doesn't really achieve the vision that you have in mind for it. So it is sort of a balancing act between those two factors, and we do try to dedicate a good amount of resources to both some tasks from both of those buckets at all times. Christine (22:39): This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 8 of 2 Merrill, you have a pretty amazing list of early angels that really took a bet on you and Tomas and Greg in the early days. Tell us a little bit more about how you decided who to bled onto the cap table and also how you balance all the advice that you were probably getting in the early days from all these different investors as well. Merrill (23:00): Yeah, yeah. I think we've been really lucky in that I think at every step of the way we've been able to really be intentional about who we bring onto the cap table and who we want to work with. And that started from the seed round when we were just basically three founders and a deck, not even really a product at that point, but we're fortunate to involve some great leaders in developer tooling from the early days, like the former CMO of a GitHub and GitLab, one of the founders of sourcegraph, PagerDuty, great developer, iconic developer tools, companies like that. And then similarly, that was back before we were building graphite then going into the series A. So our seed was Hunter Wa and Satcha Patel from Home Brew led that round Going into the a, I think we really wanted to basically put together the All- star team of everybody who's thought about code review before. (24:00): So Peter Levine, the partner from A 16 Z who joined the board was the one who did GitHub series A back in 2013 and basically came back to work with us on building the new generation of code review platforms. We also brought on GitHub's, former CRO one of their CTOs, some of their old CEOs, basically a lot of their executive team, the current GitHub, CEO and COO are investors as well, really trying to build great. And as well as that some of the great taste makers, I'd say like Kari from Linear comes to mind, Sam from Planet Scale, folks that really care deeply about crafting this new generation of well- designed developer tools and they get that ethos. And now I think going into the series B, I think similarly wanted to find partners that cared about and that really understood that vision. And I know you guys have backed some of those iconic companies like Linear and Versal. (25:05): I think that was one thing that really excited us is working with a partner that understands that value and the vision investing in the platform that we're building. We also were able to then bring on some of our largest and most helpful customers. So Shopify invested as part of the round Figma as well as one of our larger customers. We also worked with Menlo and Anthropics anthology fund. We've been working really closely with the Anthropic team on building Diamond and really being able to unlock a lot of the capabilities of their models for code review. So having those partnerships was incredibly valuable for us, and that combined with the support of some of our existing investors, like the general partnership really kind of solidified the round for us. Christine (25:56): Amazing. How did you know that it was time to go and raise a series B? How did you have confidence that it was going to be a successful raise? Merrill (26:02): Yeah, I think that it was two factors combined. One was we were really starting to see the enterprise motion work. So we'd hired our first two AEs over the summer and really had started to see them closing repeated wall-to-wall deals, getting those stood up successfully and just seeing the inbound interest growing every single day. And then the second piece was just the level of excitement around our AI code review product and seeing the potential there, seeing how quickly that technology was This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 9 of 2 improving and also how quickly how much this problem of AI generated the volume and nature of AI generated code becoming prevalent. That really gave us the confidence to say the product is inflecting. The enterprise story is there, and we have these incredible tailwinds of we know that we're going to be in the right place with the right product as the industry gets there. And we're already seeing that play out in the months since the round closed. You Christine (27:03): Started this company pre cogen and pre AI kind of really transforming software development. What happens and what do you do when there's such a big technology shift? Merrill (27:17): Yeah, it's been one of the most interesting parts of building this company, I think is just seeing how much software engineering has evolved and really living through and continuing to live through the seismic shift in how engineering is done. That's been enabled by LLMs, and I think we, in the beginning it was very easy for a lot of folks that we talked to would ask us, is software engineering just going to go away entirely? Will the machines write the code and review the code and is this even something that's still valuable? And what's been interesting is that the reality has actually been that it's made what we're doing all the more important in that I think there's been so much innovation and so much attention paid to this inner loop of how code is generated. There are amazing tools now like copilot cursor, windsurf and all these agentic development tools that help engineers write more code than ever. (28:18): But I think the reality of anyone who's been a professional software engineer knows that that's only the first part of the story for getting a code change to production. You then need to make sure that all of that is reviewed and tested and deployed both safely and efficiently. There's a lot of collaboration that goes on there and that outer loop of review, testing and deployment, that's a huge part of how engineers spend their day. And if you have 3, 5, 10 times more code that's now being generated and you're trying to then push that through the same set of rails and processes as you've had before, things start to break and you don't really see a lot of the productivity gains or the potential that's really in some of these tools because everything is just like bottlenecked in the outer loop and talking to customers. That's a lot of the reality that they're seeing now and a lot of how some of the deployments of these tools just haven't lived up to the productivity, the promise that they've had in terms of productivity gains. (29:26): And I think a lot of that is due to needing a new outer loop tool chain that can keep up with it. And that I think has been really interesting for us because everything that we were doing before AI was about accelerating the outer loop. Our whole thesis was companies like Meta and Google have figured this out, and they've built great tooling to help accelerate that coordination process to help teams that are working tens of thousands of engineers across time zones, all working in one monorepo and still managing to ship faster than any team has before. Basically, they built that tool chain to work at that scale, and now that's just more important and more valuable than ever. And so we started from even before this big shift happened, I think we were building something that we were lucky that it became all the more valuable now with the amount of code being generated. (30:26): And I think the other piece is that a lot of our customer relationships really gave us help us build the trust to now introduce AI in ways that can help to accelerate that outer loop even more. So a lot of our This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 10 of 2 customers that were using us for code review for our merge queue proactively came to us even before we started building our AI code review agent. They proactively came to us and asked, Hey, are you guys doing something here? Is there an opportunity for you to use AI to help us accelerate code review? And we'd been playing around with it for a lot internally. Really, it was when Claude 3.5 Sonic came out, that was the first moment where we resurrected what we built with the new models and we said, wow, this is actually pretty good, is now at the point where we're starting to get value out of it. (31:19): It's not quite there yet, but we can see how this is within striking distance and it's worth us investing in. And we have one of our most senior engineers who use Architected like so many of our systems and has contributed massively to the company is this engineer Dave, and he's been in the industry for 20 plus years, been a staff engineer forever, but he's one of the more AI skeptical engineers on our team. So we had what we called the Dave Test where we were basically, we said internally, once we can get this to the point where Dave would be upset if we turn this off or he really thinks that this is adding value, that's when we know, okay, this is ready to go out to customers. And a few iterations and a lot of work went into clearing that bar and that was when we first started to when we said, okay, this is ready to ship to customers and get in the hands of our actual users. Christine (32:15): I love it. The Dave test, we got to make sure that's copyrighted because that's a pretty good proxy. Well, you just launched Diamond a couple days ago now, so it obviously passed the Dave test. What else gave you confidence that it was ready? Merrill (32:31): Yeah, I think a lot of it has been testing with our existing users. So actually we rolled out the first version of this to customers in the fall. At that point, it was only an add-on that you could only buy it as an add- on to the core graphite code review platform. And the idea is pretty much the same that you basically are, our AI reviewer will give feedback on every code change in a few seconds. It'll scan for things like bugs knits, typos, code style guide inconsistencies. It's aware of your entire code base so it can tell you if something matches or doesn't match the patterns that you're using. Even security vulnerabilities, it's good at finding a lot of the oasp top 10. It's very good at doing a first pass and finding a lot of those things that a human reviewer would look for. (33:25): And the advantage here is that you can imagine in an extreme example, you have a coworker that's halfway around the world, you put up your pull request and then you have to wait a whole business day for them to wake up, give you feedback, then you wake up, you get those changes, you update your PR re-request review, and it just goes back and forth for days potentially versus within a few seconds of putting up your code change, you have that first pass all of the little things that you need to change. And so not only does it cut down at least one of those back and forth review cycles, it also then lets your colleague focus on the high level and is this the right functionality? Is this architected in the right way? Is this performant? Questions like that that are more of the focus, more of of engineering at a high level versus looking for little bugs in this and really makes reviewing a lot more focused and hopefully enjoyable as well. Christine (34:28): This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 11 of 2 Yeah, I love it. I love it. Well, you have graphite core, now you have diamonds and you're selling both of those and they seem to be being adopted pretty rapidly. What is the ROI that some of these companies are seeing from using both products and how do you quantify that in sales conversations? Merrill (34:46): Yeah, it's a great question and it's something that we really spend a lot of time in data science on and we can really tell what the differences between graphite and non graphite users at companies and even from when they adopt graphite, what the delta is in terms of engineering productivity. And oftentimes it's about 30% of faster, 30% of time save from PRS opened to merged across many of our enterprise customers. Asana was saving, we calculated their engineers were saving seven hours per week that they would've been waiting in code review, waiting for somebody else to get back to them just as a function of using graphite. And we've seen that repeat pretty much across the board. And then with Diamond, similarly, we can see how long is your average review cycle, how many meaningful issues have you found has Diamond been able find in seconds versus hours? (35:39): And then how many of those cycles we can cut down. And in many cases, we are seeing that it's saving at least one cycle time on many of the poor requests that it's commenting on. And that piece has been really valuable for us in talking to customers about the value here. And really it all comes down. One of the things we're seeing is that it's helping teams just be able to do more. It gives engineers time back that they can be building, they can be unblocked, they're able to just ship faster and not have to worry about as much of the overhead of code review. And I think that's something that you can quantify it, you can quantify it in a sales conversation, but it's also just one of those feel things of it just makes the job of engineering just feel so much better as someone who's using this every day. Christine (36:29): Yeah, that's amazing and amazing time savings, especially in a time when every organization is trying to figure out how to do more with less and go faster to keep up with everything that's happening. Merrill (36:38): Yeah, absolutely. In Christine (36:40): The competitive landscape, you mentioned Asana, you mentioned Shopify, you're working with publicly traded companies and enterprises, which is really rare at the stage of company you're at today. How do you think you've been able to go up market so quickly? What's resonated with these larger enterprises? Merrill (36:56): Yeah, I think that what's been really interesting about Graphite is that although we started from an individual developer tool, really the value compounds as teams get bigger, and as you have distributed teams, as you have more engineers, I think so much of the value really starts to compound at the scale of hundreds or even thousands of engineers working together at once. I think all those teams are now looking for all those teams are looking for how do they do more? How do they do more with less? How do they use new AI tools to help accelerate them even faster and how do they just keep up with what their competitors are doing? And not only that, I think one thing that we've seen as we've scaled is that every large enterprise has an internal developer, has some sort of internal developer productivity team, This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 12 of 2 and they have some of their best engineers often that are working on stringing together some combination of off the shelf and homegrown tools just to try to accelerate their team as much as possible. And for them, having something like graphite, that's our whole team is focused on building this not just for one company, but for so many of the world's greatest engineering organizations. That's just a huge lever for them, and it is something that gives them a lot of time back to focus on something that's more critical to their business. Christine (38:23): Yeah, that's amazing. So you have graphic core, you have diamonds, I have no doubt that there's more in store. What can you share about what we should expect from the graphite team in 2025 and beyond? Merrill (38:35): Yeah, I think you're going to see us make really meaningful investments across both of those surfaces for us. So on graphite core, we're working really closely with our enterprise partners to build the best code review and merge queue available on the market. Really going deep on that piece and thinking about how we can connect that whole experience with AI and create agents that make pull requests in motion by default instead of at rest by default today, you have to have a human who's clicking a button at every step of the way to create the PR to review it, to re-request review, to Rerun ci. And all of this we think can really be automated and kept going by default with the right AI tools. So a lot of investments in around that thesis on Diamond really making code review much more agentic, being able to have a conversation with it, have it be able to respond to feedback and stack on those changes that you want done and really have this be a companion that's moving prs through that process end to end, really going to go deep on what that user experience looks like. (39:50): And especially as more, I think as development shifts more from being an activity of writing code to one of reviewing AI generated code, I think that experience is going to become probably the most valuable surface in the software engineering life lifecycle. And we really want to build the best platform where that activity takes place and kind of be the home screen that developers are living in pretty much all their professional lives. Christine (40:18): Amazing. You talk about your customers and some of the competitive pressures that they're facing. You're also building an AI dev tool startup, and we're in San Francisco today, but you're based in New York and you walk around San Francisco and there's a lot of people shouting from the rooftops. You have to be in the Bay Area right now. You have to be in office 5, 6, 7 days a week to really be staying ahead of the curve. You're doing things a little differently. What's your take on all that? Merrill (40:43): Yeah, I think from the beginning we had a thesis that New York was emerging as one of the major technology hubs, and there was a lot there going to be more great engineering talent that was shifting from the bay and from some of the other places that had the diaspora during Covid to New York. And I think that that's really played out for us. It's also made it much easier. We're also five days a week in office. I think New York as a city is just designed for working in office. Our apartments are small, there's great public transportation. It is the best city in America, I think, to build five days a week in office. And I think that's helped us a lot. The other piece I'd say is there is an advantage provided that there's a great This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 13 of 2 talent pool in New York. There's an advantage to being the big fish in a slightly smaller pond if you want to build a great dev tool startup in New York. (41:37): There are only a few that really come to mind, and I think Graphite is building one of, if not the defining New York dev tool startup from this cohort. And that really, I think being able to kind of wholly own that category and really attract the top engineering talent in New York that wants to build amazing dev tools is something that really works to our advantage. And we've actually relocated some engineers from the Bay to New York. We sponsor that. For anyone that wants to move and join us in New York, just make it as easy as possible. And I think that's really worked to our advantage in many ways. And obviously a lot of our customers are here. We do a lot of events and community engagement out here as well. But building in New York I think has been one of the best decisions that we made early on. Christine (42:24): Yeah, amazing. Well, being in New York, there's also a lot of other things to do outside of technology, and I know you keep yourself busy. What do you do outside of the office? How do you stay staying while building a fast moving and fast growing startup? Merrill (42:40): Yeah, I think that it's important to have some amount of balance or harmony in your life where obviously building a company can be all consuming, and I think that having something athletic that you focus on and something creative that can kind of be an outlet that's different from the day-to-day work of building a company really helps me stay energized and stay excited about building this company for the long, long term. And I run pretty competitively. I started doing half marathons and marathons recently. I'm trying to qualify. Christine (43:19): I think you're underselling yourself. How fast was your last marathon, Merrill? Merrill (43:23): 2 55. Christine (43:24): 2 55. That is insane. That is insane. And you're going to do qualify for Boston soon? Merrill (43:31): Yeah, Christine (43:31): That's the next goal. What's the next marathon? Merrill (43:32): That's the next goal. Grandma's marathon in Minnesota in June. Christine (43:35): This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 14 of 2 Wow. So we need a break. 2 53. Merrill (43:38): 2 53. Yeah. Christine (43:39): Okay. Merrill (43:39): Yeah. Christine (43:40): You believe? Merrill (43:40): Yeah, Christine (43:41): I believe, Merrill (43:41): Yeah, I think it's going to happen. Christine (43:42): Okay. So you do running. Merrill (43:44): Yep. Christine (43:44): What else? Merrill (43:46): I haven't been doing this as much, but I also DJ a bit for fun on the side. I love electronic music and I've been DJing at clubs in New York for some time on the side as well. That's definitely a fun, creative outlet and something that gives me just a totally different type of thinking and challenge than building a company. But in some ways though, I think that the feedback loop is sort of similar. It's like with building a company, there's the same idea of putting something out there, getting feedback, iterating on it really quickly. And I joke sometimes that DJing is kind of like that too, where it's like you play something, you see how people react to it, you iterate on it, you mix things together and you kind of find the right blend. And I think that tight feedback loop and just the excitement when you really get it right with either a mix or a new product launch, I think is the best feeling in the world. Christine (44:46): This transcript was exported on Mar 27, 2025 - view latest version here. Accel - Season 3 - Graphite - Episode - Fine Cut (Completed 03/27/25) Transcript by Rev.com Page 15 of 2 What advice do you have for young entrepreneurs just starting out and building their companies? Merrill (44:51): I'd say find people that you really love working with, really pay attention to who are the class project partners that you really mesh with, well, your coworkers at your first few roles, those are going to be your future. I think the idea of founder dating is probably worse than actual dating in many ways. It's so hard to build that foundation of trust in retrospect. You really need to have a deep friendship and foundation of working together for a long time because when it gets hard, that's the only thing that you have to rely on. And so many startups die of suicide instead of hunger, and much of that is due to founder disagreements and having, so I think that building those relationships and really paying attention to that early on is one of the most valuable things you can do. Second to that is also finding a problem area that you're really interested in and learning. Who are the people that are thinking about this? What are their problems? Just obsessing over that as opposed to, you can't really start from the technology. You really have to start from, is this a problem I care about and can I see myself working? Because you're going to spend the next 10 years of your life if things go well talking to those customers every single day. And unless you're really excited about building something that they love using, you're just not going to be able to create an enduring company there. Christine (46:17): That's amazing. Well, thank you so much for being here today. Congrats on everything. Merrill (46:21): Thank you for having me Christine (46:22): Light wise, and it's going to be a massive success. Merrill (46:24): Yeah, I was real excited for the year ahead, and there's so much that even that we have in the works right now in terms of our AI code review platform and just our core code review and merge tooling that we can't wait to show customers. So Christine (46:40): Amazing. Well, we're all going to be on the edge of our seats for all the new graphite product launches. Good luck. Merrill (46:45): Awesome. Thank you. Thanks. ‍ --- [View episode](https://accel.com/podcast-episodes/graphites-merrill-lutsky-on-revolutionizing-code-review-for-the-ai-era) [View all content](https://accel.com/sitemap.md)