Boris Cherny: Claude Code & the Future of Engineering | Acquired Unplugged presented by WorkOS

Ben Gilbert 0:00 All right, so we need the origin story. Where did Claude Code come from?

Boris Cherny 0:05 By the way, it's, it's, it's crazy being here and hearing your guys' voices because I'm used to hearing it, like, from here.

Ben Gilbert 0:10 And probably at 2X.

Boris Cherny 0:11 [laughs] Probably at 2X. [laughs]

David Rosenthal 0:14 2X and with all the, um, editing that we do on all of our little verbal tics and-

Boris Cherny 0:19 Yeah

David Rosenthal 0:19 ... you know.

Boris Cherny 0:19 All that. Um, yeah, it was, it was sort of an accident in, in a lot of ways. Um, so I, I joined a team at Anthropic late in 2024, and it, it was kind of a prototyping team. Um, it was called the Labs team. We actually brought it back. And the idea was to figure out what's the next big product. And at the same time, we push the frontier of the model, and we figure out how do we improve the model to support that best, you know, that future product that hasn't been built. And it, it's just hard to know which way to evolve the model-

Ben Gilbert 0:49 Hm

Boris Cherny 0:50 ... because until there's a product that really pushes the frontier, then you don't really know. And at the time there was this, like, just crazy feeling of product overhang that, you know, the model can do all the stuff that the... There's no product that has yet captured. Um, you know, i- if you think about coding tools at the time, there was, like, auto-complete tools. Um, there was, like, you can ask the agent a question, but it couldn't write code.

David Rosenthal 1:13 Yeah.

Boris Cherny 1:14 And it's because-

David Rosenthal 1:14 Copilot and, yeah.

Boris Cherny 1:15 Yeah, and, like, until recently, the model just wasn't good. It w- it couldn't really do it. And so we felt like there, there's room to just go all in, build a product that's just a coding agent. And, um, you know, we built it and it kinda sucked-

David Rosenthal 1:27 [laughs]

Boris Cherny 1:28 ... at the beginning. It, it wrote maybe, you know, like, 10, 20% of my code for-

David Rosenthal 1:33 Well-

Boris Cherny 1:33 ... for a while.

David Rosenthal 1:33 Real quick, before, uh, before this moment, what was the state of coding at Anthropic? How much of a focus was it?

Boris Cherny 1:41 So, you know, uh, internally at Anthropic, we were using IDEs still at, at the time. Um, but for Anthropic as a company, we've always cared about coding. Um, we've always cared about tool use. We've always cared about computer use. This has always been kind of the core research direction because, uh, you know, for us, we exist to research AI safety. That's the reason that we exist. Like, if, if you ask, like, a random person at Anthropic, you just, like, pull them over in the hallway and ask, "Hey, why are you here?" They're gonna say, "AI safety." That's the reason every single person is at the company. That's... Th- they deeply believe, including me, this is just the single most important problem to solve. And there's kind of a bunch of ways to solve it. You can look at kind of mech- mechanistic interpretability, you can do alignment work. There's all sorts of ways to study the model in a petri dish. But fundamentally, you have to study it in the wild to see what it does once it's safe in all these other ways, and so that's kinda where Claude Code comes in.

Ben Gilbert 2:34 Hm.

Boris Cherny 2:34 But, you know, for Anthropic as a company, the direction has always been safety. And the way you study safety, you know, it has all these kind of layers, including studying the model in the wild. And, uh, kinda what's a useful way to do that? It's coding because that's the way the model interacts with the world. And so if you wanna study kind of various sorts of model misalignment, you wanna, uh, make it useful enough that people use it so that you can study it, coding is just a very obvious application. Um, so that's just from the beginning, that's, that's been the focus.

Ben Gilbert 3:03 It's just such an unbelievably clean universe 'cause you've got all this training data. It either works or doesn't work, and, like, depending on the language, it either compiles or doesn't compile or it either... You know, there's sort of a very clear, like, pass/fail sort of ab- ability to test it. Um, and it's a very constrained universe of correct solutions. English is-- There's an infinite number of correct, beautiful poems, but there's a very constrained, uh, there's a non-infinite number of ways that you can write the correct code to solve a certain problem.

Boris Cherny 3:38 Yeah, that's right.

Ben Gilbert 3:38 And it's, there's sort of a, an elegance to using it as your petri dish given that.

Boris Cherny 3:43 Yeah, that's right. There, there was a... You know, you know Hofstadter? He's the guy that wrote, uh, Godel, Escher, Bach. Did anyone read that?

Ben Gilbert 3:48 Oh, yeah.

David Rosenthal 3:48 Oh, yeah, yeah. Yeah. Great book.

Boris Cherny 3:50 Yeah. Yeah, it's really good. But he, he wrote this other really weird book in the, in the '80s about, uh, like, his idea about, like, the path to artificial intelligence, 'cause he's a, like, he's a professor. He taught AI in school and, like, he was kinda teaching at the time. And he was, he was talking about, like, the, the, the basis of cognition, the basis of intelligence is recognizing patterns.

David Rosenthal 4:07 Hm.

Boris Cherny 4:08 And one way that he studied that is one of his, one of his hobbies was translating poetry from English to French. And a direct translation is rarely the right translation. There's actually a lot of taste that goes into it, and sometimes you wanna, like, fudge it a little bit so that, like, the meter is good, and there's kind of various ways to do it. So yeah, it's like a fuzzy problem.

Ben Gilbert 4:24 Hm.

Boris Cherny 4:25 But coding is easier to solve, I think. It's also very commercially valuable, um, so it helps us kinda build a business, and it's also the kind of business that our favorite customers, enterprises, startups, companies have. And so it helps us have this kinda aligned business model so, you know, we don't need to do ads and, you know, we can kind of focus on safety, which is the thing that we care about. It helps us build a business where we can actually do that.

Ben Gilbert 4:50 Hm.

David Rosenthal 4:51 Um, you came from Meta before coming to Anthropic, uh, and you had a lot of focus on, on code at Meta. Uh, I, I'm curious, connecting the dots looking back now, um, how did what you were doing at Meta help inform the beginnings of Claude Code?

Boris Cherny 5:12 Hm. For me, building dev tools has always been a side hustle. It was never the main thing.

David Rosenthal 5:19 Hm.

Boris Cherny 5:19 And I feel like that's the m- that's the best way to build dev tools because you wanna focus on solving a business problem and building a thing that people love and that's useful for people. That's fundamentally what we're here for as engineers of it. You know, obvious- YC, like, beats this into you.

David Rosenthal 5:32 [laughs]

Boris Cherny 5:33 Unless I, I, I feel this way.

David Rosenthal 5:34 Had you done YC before? [laughs]

Boris Cherny 5:36 Yeah, yeah. I was actually, I, I was the first hire at a YC company. This was back in, like, 2010, 2011.

Ben Gilbert 5:43 Hm.

Boris Cherny 5:43 Something like that. It was, like, one of the first few batches.

David Rosenthal 5:46 Wow, cool.

Ben Gilbert 5:46 [laughs]

Boris Cherny 5:47 Um, so for me, dev tools has always been this kinda side hustle, and I, I build product, and on the way to building product, if, uh, you know, the dev X isn't good, then I'll build tools to make that a little bit better. That's just always been the way that I approach it. Um, and, and so here I just applied kinda that mindset of build the thing for yourself first, and then hopefully it's useful for others. And, uh, I was surprised to see that it is, which is awesome.

Ben Gilbert 6:09 So you said it was kinda crappy at first, and you only trusted it with, like, 10% of your work-

Boris Cherny 6:13 [coughs]

Ben Gilbert 6:13 ... or only did about 10% of your work. What was the bottleneck early on, and how did you solve it to, to make it better? 'Cause, uh, from what I understand, it's been, like, six months since you've written a line of code in your job.

Boris Cherny 6:24 Yes.

Ben Gilbert 6:25 So that's extremely different [laughs] than it, it only does about 10%. What changed?

Boris Cherny 6:31 Um, i- I, I remember it. It was in, in May. It was Sonnet 4 and Opus 4.

David Rosenthal 6:37 Hm.

Boris Cherny 6:37 Uh, and then in, uh, in November it was Opus 4.5.

Ben Gilbert 6:41 So it was just the mo- the underlying model changing.

Boris Cherny 6:44 It was the model. There's a lot of work that went into the harness. There's a lot of work into making Claude Code better. You know, like, everyone likes to use this stuff differently. This is the thing about building for engineers. It's not a consumer product because engineers are so opinionated in the way they like to use stuff. And so, you know, we started with a CLI, but then we built a, we built a desktop app, we built a mobile app, we built, uh, iOS and Android apps. We have, like, Slack app, GitHub app. You can use Claude Code in whatever way you want. And so a lot of this has, was us innovating, learning, trying to figure out what features are useful. And this is, like, plan mode, for example, came out of this.

David Rosenthal 7:15 Hm.

Boris Cherny 7:15 A lot of different kinda tools and experiences came out of this. But fundamentally, when I think about the step changes and what percent of my code the model wrote, it was just the model.

David Rosenthal 7:24 Hm.

Boris Cherny 7:24 Like, the model got better, and it just went up.

David Rosenthal 7:27 How tightly, uh, to the extent you are able to talk about it, how tightly was the experience that Claude Code was having feeding back into the mo- the model work leading up to 4.0, 4.5, et cetera?

Boris Cherny 7:42 So every, you know, everyone at Anthropic uses Claude Code every day. Um, so the researchers building the model use Claude Code, and the people building the product use Claude Code. So that, that's essentially the cycle.

Ben Gilbert 7:53 Do you have any metrics that, like, um, ways to numerically describe how much it's changed the company's trajectory? Like, it feels from the outside that you guys are shipping a whole lot more product now than you used to. It seems reasonable to assume a lot of that is because you guys are using Claude Code all over the company internally. Uh, what's the right way to really grasp how much of that is attributable to Claude Code?

Boris Cherny 8:17 Yeah, so be- being at a AI lab you get [laughs] you get kinda used to thinking in exponentials. Essentially, all the charts are exponential, so we actually use, like, log linear charts for pretty much everything.

David Rosenthal 8:26 [laughs]

Boris Cherny 8:27 Um, [laughs] so it's like x-axis is, you know, like, linear, and then the Y is, like, log linear.

Ben Gilbert 8:31 [laughs]

Boris Cherny 8:32 Um, and so, you know, everything's kind of exponential. Like, the revenue is exponential. The usage is exponential, um, which is just, like, the funnest problem as an engineer, and thank you for bearing with it. Um, and, uh, e- essentially, like, everything is going up including the, like, the code. Um, and so when we look at the lines of code written, the number of pull requests written by people at Anthropic since we released Claude Code, this has grown many hundreds of percentage points. Um, I think the latest stat we shared was, like, something like 3X, um, but this is actually very, very outdated, so it's, it's a lot higher now.

Ben Gilbert 9:04 What, what was 3X?

Boris Cherny 9:05 Uh, the amount of code per engineer that everyone at Anthropic is writing.

Ben Gilbert 9:10 I see.

Boris Cherny 9:11 So the size of the engineering team went up, you know, like many multiples, and usually what happens as the size of the engineering team grows is productivity goes down. And when I was at Meta, one of my responsibilities was code quality for all the code bases, and the reason we cared about that was it's useful for productivity. If the code is high quality, it makes engineers more productive, and we kinda showed this empirically. Um, and typically what happens is the code quality gets worse, productivity goes down. Another thing that happens is as the number of engineers grows, the productivity goes down because the new people are kinda asking engineers, like, "How do I do this? How do I do this?" So the engineers they're asking don't get to write code, and it takes just, you know, like, it takes ramp-up time. And the, the first thing that we saw at Anthropic is when new people join, the ramp-up time is, like, two days.

Ben Gilbert 9:55 [laughs]

Boris Cherny 9:56 'Cause, you know, like, this used to be weeks, but now it's two days 'cause you just, like, ask Claude. And we still have to explain to new people that join the team-

Ben Gilbert 10:03 Wow

Boris Cherny 10:03 ... like, people ask, like, "How do I query the database?" And the answer is open Claude, run it in the Claude Code code base, and have Claude query the database.

David Rosenthal 10:11 Yeah.

Boris Cherny 10:11 'Cause there's, like, a skill for querying the database, and it just knows.

David Rosenthal 10:13 Wow.

Ben Gilbert 10:13 So of, of the new engineers that you've hired in 2026, how many of them have written any code?

Boris Cherny 10:21 Um, I mean, define writing code. Uh-

David Rosenthal 10:26 Right. Right.

Ben Gilbert 10:26 [laughs]

David Rosenthal 10:26 That's... [laughs] Wait, wait, wait. How do you define writing code? [laughs]

Boris Cherny 10:31 You know, for, for me it's like engineering has always been, like, like, you know, my grandpa coded with punch cards, like, in the Soviet Union. Like, you, like, for him, like, programming was like a, it was, like, a paper thing, and you, like, punch holes into it, and you feed into this, like, big machine, and it chugs for a while, and it gives you the answer. That was, that was coding.

Ben Gilbert 10:45 Right. It's like my dad, who wrote a lot of assembly, would've, like, made fun of me for writing Python and be like, "You're not coding." [laughs]

Boris Cherny 10:51 Exactly. Exactly. It's like it's not for, you know, real engineers or it's very high level, but this is sort of the nature of programming is the, the, uh, level of abstraction always goes up, so it's, it's gone, it went from, like, switches to punch cards to, like, assembly, then to, like, COBOL, Fortran, then Java and the, the high-level languages and then, like, JavaScript and Python, then the really high-level languages.

Ben Gilbert 11:09 Hm.

Boris Cherny 11:10 This has just always happened and, like, I see what we're doing as somewhere on this continuum. Um, and the, so for me, the way that I coded a year ago was I wrote code with some kind of auto-complete in IDE. In November, I uninstalled my IDE 'cause [laughs] I wasn't using it. [laughs] I, I just, like, I, I was just, you know, for, I, I was like, in the last month I just haven't opened it, so I, I'll, I'll just uninstall it.

David Rosenthal 11:34 Wow.

Boris Cherny 11:35 At that point I was running, you know, maybe 5, 10 Claudes in parallel, and my coding was prompting Claude to, to write code. Now it's actually leveled up, I think, again to the next level of abstraction where I don't prompt Claude anymore. I have loops that are running. They're the ones that are prompting Claude and kinda figuring out what to do. My job is to write loops. And th- this is this kinda next transition that I think we're gonna see in the next few months and may- you know, maybe through the rest of the year.

David Rosenthal 11:58 Wow.

Ben Gilbert 11:59 Hm.

David Rosenthal 11:59 Um, hypothetical me, obviously not actual me, uh, is an engineer who wants to join the engineering team at Anthropic. Um, how do you evaluate me today? [laughs]

Boris Cherny 12:15 I think something that I look for in our team in particular for Claude Code is we love generalists. So I, I think a thing that we started to see maybe six months ago is there were very few engineers on the team that were just doing engineering in the traditional sense of like, you know, you have a user researcher, they talk to the user, and then they handed that down in a document to a designer who then made a mock, and then that went to a product manager, you know, who scoped it, and then it goes to an engineer for implementation.

Ben Gilbert 12:42 I, I worked at Microsoft, I'm extremely familiar with this process.

Boris Cherny 12:45 [laughs] This is what it used to be. But it, you know, we, we just realized this is not what the team is doing anymore. And our team was always kind of more prototypey than this, but maybe six months ago, the thing we found is every engineer on the team does scoping. Everyone talks to users every day. People are doing design. Engineers, everyone on the team is comfortable, like, pulling data and doing data science work, building dashboards and things like that. So I think, like, we're, we're kinda starting to see all the roles merge and kind of melt into one builder. I think, like, Satya calls it a builder, so it's something like that.

Ben Gilbert 13:17 Yeah. It's almost like a product engineer, but you don't even really necessarily have to be an engineer.

Boris Cherny 13:23 That's right. And, you know, like we have designers that are shipping code, our finance guy ships code. Um, I think this is-

Ben Gilbert 13:30 Our chief of staff ships code.

Boris Cherny 13:33 Yeah.

Ben Gilbert 13:33 I mean, w- actually, all of our infrastructure at, at Acquired now is, is pretty rapidly shifted to Claude Code, um, which is why... Can I tell a little fun Acquired story? So we, um, we are an audio podcast. YouTube doesn't have a way to upload just audio. You have to upload something in the video. It should be interesting. Um-

David Rosenthal 13:57 For a long time it was just a, a, a black screen MP4. [laughs]

Ben Gilbert 14:01 Uh, so a transcript would be a good idea, but we don't have a transcript yet, 'cause it takes a, a week for humans to make the transcript. Fortunately, there's AI. Anyway, um, there's a whole, like, Claude Code set of scripts that I sloppily wrote... Well, I and Claude wrote, and then at some point I sort of heaved it over to David, and I was like, "Can you, like, continue working on this?" And there, at first, a little resistance, but then you, you did spin up an I- like-

David Rosenthal 14:29 Yeah. Yeah, yeah

Ben Gilbert 14:30 ... yeah, um, open the terminal.

David Rosenthal 14:32 [laughs]

Ben Gilbert 14:34 O- one day after-

David Rosenthal 14:35 Yeah, I took CS classes in college, and that was the last time I opened a terminal.

Ben Gilbert 14:38 One day after that, uh, Cowork comes out, and the, you know, the, the pitch for Cowork is it, it's if you want the power of Claude Code, but you don't wanna have to deal with the terminal and, you know, typing commands and, um, installing NPM and blah, blah, blah-

Boris Cherny 14:56 We actually heard that David had to open a terminal, so we were like, "We have to release this. We have one day." [laughs]

Ben Gilbert 15:01 So, uh, I wanna use that as my, like, hook to, like, uh, tell us the story of Claude Cowork, because to me it seems Claude Cowork is a much, much bigger market of people who don't want to open terminals than people who do know how to open terminals. It's kind of interesting that it went the direction of Claude Code first, and then to be honest, like, kind of a really long time before Claude Cowork came out.

Boris Cherny 15:26 Yeah, that's right. Honestly, it was this story. It, it was like, it was some version of this played out. Like, i- in, in April, I remember walking into the office, and right next to where the Claude Code, you know, the, a few of us sit, there's the, a couple data scientists. And I walked in, and one of the data scientists had Claude Code up on his screen, and th- this was like, this was back in the day when Claude Code was only available in a terminal. It wasn't yet available in the desktop app, it wasn't available in the mobile app, so, like, you really had to know what you were doing. And I asked Brandon, like, "Dude, what are you doing? Are you, like, dogfooding it, just trying to kind of figure out what this thing is?" 'Cause he, you know, like, he just joined the team. And he was like, "No, I'm using it for data analysis." And so he figured out how to open a terminal, he figured out how to, um, you know, download Node.js, 'cause you had to do that. Uh, he figured out how to install Claude Code, how to set up a API key. I don't think we even had subscriptions then. And he was just, like, using it to do analysis. And then the, the next week, all of the data scientists had multiple Claude Code windows-

David Rosenthal 16:27 Yeah [laughs]

Boris Cherny 16:27 ... doing their analysis, 'cause it was just good at it. Um, and then fast-forward to maybe May or June or something, and there, there was some guy on Twitter... Do you guys see this? He was using it to grow his tomato plants. He has like, he had, like, a little webcam, like he set it up to, like, just monitor his tomato plants, and it was controlling, like, the nutrition for the plants and, like, over, over weeks. And then at some point, like, the little tomato, like, butted, and Claude was like, "This is so delightful."

Ben Gilbert 16:53 [laughs]

Boris Cherny 16:53 Like, all our work has paid off. And, uh, like, I just saw this tweet and I was like, "Oh my God, I think it's time." Like, now we're, we're starting to break through to the mainstream. It took a while to get engineers to understand what this is, and finally they did, and now we're starting to see a critical mass of non-engineers do it. And when the weight and demand is that strong, I think that just means it's like, it's the time to build a product. And so we started exploring. There was a bunch of different ideas. Cowork itself we built in, it was, like, a little over a week, like maybe eight, nine days.

Ben Gilbert 17:23 Hmm.

Boris Cherny 17:23 Something like that. It was 100% built using Claude Code. It was one of a bunch of ideas, and as soon as we had it, we were like, "This is it."

Ben Gilbert 17:29 What were the other paths that you killed? 'Cause it, um... If I were to describe the sort of creative brief as take all the power of Claude Code and make it available to people who don't wanna open a terminal, it doesn't really translate one-to-one. You kinda have to be opinionated in the product that you build to bite off some set of the work that Claude Code can do.

Boris Cherny 17:52 Yeah, we tried... Man, I don't even remember all the prototypes. We tried something, I think, that was based in Slack, and it didn't really work, 'cause building Slack chatbots that feel good is just really hard.

Ben Gilbert 18:01 Hmm.

Boris Cherny 18:02 Um, we tried something that was based on web. There was a bunch of different web-based prototypes, um, that we weren't sure about, and none of them really felt great. Like, there, there's something about being in a browser that just didn't feel really good. Um, and it, it also, if it's in a browser, it doesn't have access to all your tools. So I have a Word document on my desktop, I can't tell something in the browser-

Ben Gilbert 18:19 Right. File system access feels-

Boris Cherny 18:21 ... necessary. And just that little bit of friction of, like, having to drag and drop the file to the, to the browser, that was enough to make it feel not good. And I, I think this is just back to the way that we think of our product is we build the thing that we love. And you wanna build a product that you yourself fall in love with and that you use every day, and then maybe some users will like it too, and hopefully it's useful for other people too.

Ben Gilbert 18:40 Hmm. I have, um, it has been surprising to me how my Claude Code usage has dropped because of how much Python gets written in Claude, the chat window. I'll ask it questions that require a lot of data analysis, and I can see, like, oh, it's, it's writing some software right now to do the analysis, and then it's gonna write a bunch of ATM- HTML and CSS and JavaScript, and then render a brand new UI for me to play with it in. And that used a lot of Claude code generation, although I never once interacted with Claude Code the product to, to have that be accomplished.

Boris Cherny 19:14 Yeah, yeah. There, there's this kinda interesting kinda theoretical thing about, like, when do you wanna do the computation? Do you wanna do it in real time when you sample from the model and kinda do inference? Or do you wanna kinda pre-compute so you can have the model write a program, and then you can run that program repeatedly and that's free for you? And essentially we wanna do the latter as much as possible, um, but it really it's up to the model. You know, like, th- this is just back to, like, the model exists as software and it needs a way to interact with the world. Because it exists as software, the way it does this is, is coding. That's just kinda the natural language it speaks. It's the thing it understands.

Ben Gilbert 19:45 Hmm.

Boris Cherny 19:45 And so, you know, like, from the earliest days of Claude Code, I- the very first version I gave it a batch tool so it can run commands on my computer. I didn't tell it anything about the tool, and it just kinda figured out how to use it. And, you know, there's this anecdote I've told 1,000 times, but I asked it, like, what music am I listening to? And it opened up the batch tool, it wrote a little program that wrote a AppleScript script, and, like, I've never written AppleScript.

Ben Gilbert 20:08 [laughs]

Boris Cherny 20:09 And it wrote a little AppleScript, and it opened my music player and was like, "This is the song that you're listening to." And even back then, this was, like, Sonnet 3.5. That was, like, not a very intelligent model. It was, like, the best at the time, but, you know, by now- by modern standards it's not very intelligent. Um, it, it just knew how to do this. So there, there's something about writing code, there's something about, you know, using tools to interact with the world that's- it's just something the model wants to do.

Ben Gilbert 20:31 Hmm. I wanna ask, I'm gonna come off of this product-specific questions for a minute to ask a little bit about culture and org design. Um, you're a member of the technical staff. I think a lot of people have that title.

Boris Cherny 20:46 Hmm. [laughs]

Ben Gilbert 20:48 Um, what's up with that? What are the pros? What are the cons?

Boris Cherny 20:52 [laughs]

Ben Gilbert 20:52 Would you recommend that to this group?

Boris Cherny 20:55 Okay, so the, so the worst thing th- that was pretty annoying when I joined is, uh, you message someone on Slack and, you know, it just says member of technical staff. And so you're like, is this person, like, a designer or are they an engineer? [laughs] Like, manager? You don't know.

Ben Gilbert 21:07 And, like, what do they work on?

Boris Cherny 21:09 What do they work on? Yeah, you don't know. Um, I actually really like it. Um, when I, when I was at Meta something I really liked is every software engineer, their title was software engineer. There's no, like, senior software engineer or principal software engineer.

Ben Gilbert 21:21 Oh, really?

Boris Cherny 21:21 Yeah, it's, like, not a thing.

Ben Gilbert 21:22 Huh.

Boris Cherny 21:23 Um, and I, I, I like it because actually if you give people kinda senior titles, sometimes they'll give you bad ideas, and people out of deference will just go with those ideas.

Ben Gilbert 21:31 Hmm.

Boris Cherny 21:32 But actually they should push back. And it's actually a really good cultural forcing function to put everyone on the same playing field in that way.

Ben Gilbert 21:37 Does it come out though? Like, "Okay, yeah, I know you're just a software engineer and I'm just a software engineer, but, like, I'm well aware you're a level something software engineer even though it's not in your title."

Boris Cherny 21:47 I think sometimes you know this, but oftentimes you don't.

Ben Gilbert 21:49 Hmm.

Boris Cherny 21:49 Yeah. Like, um, you know, I was like a... When I was at Facebook I w- when I joined I was, like, a L4 engineer, and I had an idea and then I, like, I found, like, the VP of connectivity and, like, went to him and was like, "Here's my idea. Let's go build it." And, like, he didn't know what level I am.

Ben Gilbert 22:03 Oh, that's interesting.

Boris Cherny 22:04 And, like, he, it was a bad idea and it didn't work.

Ben Gilbert 22:07 [laughs]

Boris Cherny 22:08 But then I did this again with a different VP, and then, you know, that also didn't work, but, but the third time it worked.

Ben Gilbert 22:12 [laughs]

Boris Cherny 22:12 And then, you know, we, we started a team and we started building stuff, and then moved on to the next idea. I, and, and I kinda see this all the time now. You know, people on our team, old, these kinda old seniority ideas, it's, it's sort of like it doesn't matter anymore in a, in a lot of ways. Like, these engineers that have 20 years of experience, 30 year, like, they have to unlearn so much, and you spend months teaching them how to unlearn all these old habits that just aren't relevant anymore. Whereas sometimes a new grad joins the team, and they will teach me something about how to use Claude Code better.

Ben Gilbert 22:43 Hmm.

Boris Cherny 22:43 Because they think about it natively, whereas it's not my instinct. And it, it's extra hard because with every new model you just have to recalibrate and relearn. But going back to the member of technical staff thing, I, I think it's actually a really important thing because all these old distinctions about engineer versus PM versus designer versus user researcher, by the end of the year it's gone. And so this helps us kind of very clearly see this and kinda act in this way today. And this is, I think, a, a thing you have to do within a AI, AI lab is kinda push everyone to be more agile and think about the future more in this way.

Ben Gilbert 23:19 Hmm.

David Rosenthal 23:20 I'm curious on that front, maybe generalize a bit, you know, for all the founders, all the companies in this room, um, what should they be thinking about now for their organizations by the end of the year? Um, what-

Boris Cherny 23:38 What-

David Rosenthal 23:39 ... do you think that founders and companies are gonna have to change in their mindset-

Boris Cherny 23:43 Yeah

David Rosenthal 23:43 ... over the coming set of months?

Boris Cherny 23:45 Hmm. I would, I would think today the thing that I would do is I would give everyone as many, and I'm, I'm gonna say this 'cause I work at a AI lab of course, but give everyone as many tokens as possible.

David Rosenthal 23:55 [laughs]

Ben Gilbert 23:58 The more... To, to quote Jensen, "The more you buy, the more you save."

David Rosenthal 24:00 The more you buy, the more you save. [laughs]

Ben Gilbert 24:04 [laughs]

Boris Cherny 24:05 So this is, this is sort of the place that I would start is, like, give people as many tokens as possible, let them experiment. And the other thing I would do is underfund everything a little bit. So if you're working on a project and you're like, "This feels like it needs, like, four engineers," put two engineers on it and give them a bunch of tokens and have them figure out how to do it, and chances are they can do it There's probably a bunch of stuff they can automate, there's a bunch of stuff they can streamline. And because they're automating it, they can do it better next time. It'll just be cheaper next time. And of course, like if-

David Rosenthal 24:32 There's like a compounding effect to-

Boris Cherny 24:34 It's compounding

David Rosenthal 24:35 ... a compounding effect to having fewer resources besides tokens.

Boris Cherny 24:39 That's right. It's sort of like, you know, like i- i- in business and product, y- you- you make so many decisions, it's really good n- to have a set of principles so you don't have to... You can lean on the principles and you don't have to make ad hoc decisions every time. And oftentimes founders and, you know, companies, they have these, like, principles documents. And so this is kind of like that. But it, you know, the model can use these principles also, and, you know, it comes in the form of skills. And, and of course, like, if there's some use case that takes off and uses a bunch of tokens, that's when you go in and optimize it and try to make it efficient.

Ben Gilbert 25:07 But in general, if I were to parrot back the advice to you of, um, staff with fewer humans, r- shift budget from humans to tokens, what the output of that is is to raise your upfront costs but decrease your ongoing costs 'cause you- you've already, like... It's almost like pre-compiling. Like, you've done a big pile of work up front so that the over and over and over again tasks are easy and streamlined.

Boris Cherny 25:34 Yeah, yeah. That- that's maybe a way to think about it. And, and because you put less people on it, that means, you know, the people at your company can do much more.

Ben Gilbert 25:41 This feels like the biggest reckoning to me of building teams, is people are pretty used to having a title and a discipline, and people take great pride in being a good, a good PM. I'm gonna write a bunch of blog posts and stuff about the, the discipline of product. You know, there's a lot of those people out there. Or d- designers, I can't wait to show off my portfolio of my beautiful designs. And, like, do we all just need to, within 12 months, shake off the idea that we are an anything, and everybody is just this, like, flexible bag of token generation?

David Rosenthal 26:17 [laughs]

Boris Cherny 26:18 [laughs]

David Rosenthal 26:18 Wow, Ben, you're so optimistic.

Boris Cherny 26:20 [laughs] Uh, I, I, I might use slightly different words, but yeah. [laughs]

Ben Gilbert 26:24 Yeah. [laughs]

Boris Cherny 26:25 Something like that.

Ben Gilbert 26:26 You should make a T-shirt out of that.

Boris Cherny 26:27 [laughs] I think, I think it's something like that. Uh, so, you know, for, for me, I, I've been an engineer for, for a bit, but I, I've always worn different hats. Like, I, I've started startups, startups before and I've been on the business side. I've worked on the product side, I've been on the user research side, on the design side. So for me personally, I just love building product. I don't, I don't care that it has to be an engineering hat that I'm wearing. And so it, it's sort of like right now I see that being borne out. So yeah, I feel like right now this is just the golden age of, of the generalist. People that want to do more than one thing, it's just, it's never been more fun. It's never been easier.

Ben Gilbert 26:59 Hmm.

David Rosenthal 27:01 Maybe one more question, um, uh, we spoke about with Garrett a little bit, I'm curious your take on it. Um, how do you think about taste, uh, either you personally or Anthropic?

Boris Cherny 27:15 I, I feel like every time I think there's, you know, something s- kind of special about the way that I do coding, I'm wrong. I, I used to have all these opinions about the way that we write the code. Like, um, you know, I, I, I really like functional programming, and so, you know, I'm a big, like, yeah, I like, like, Haskell and Scala and these kind of weird functional languages. And at the very beginning, I had this rule for our code base, we have no classes in the code base. We only have functions in the code base. 'Cause that's the way that I write my code. That's the way that I like it. And, um, on the weekends, engineers just started sneaking in changes that had classes.

David Rosenthal 27:48 Hmm.

Boris Cherny 27:49 And, uh, I would find that on Monday, and then I would, like, pull out the pull request, and I would be like, "No, no, no, don't do that."

David Rosenthal 27:53 [laughs]

Boris Cherny 27:54 And then at some point the models started writing all the code and they just started writing classes, and I was like, "Okay, well maybe the model's right." Like, maybe this opinion's kind of stupid-

David Rosenthal 28:01 [laughs]

Boris Cherny 28:01 ... and it's just, like, this particular thing that I have, but, and then it doesn't matter 'cause the business outcome is met.

David Rosenthal 28:05 Hmm.

Boris Cherny 28:06 And it's met faster as a result. And the code actually is not bad. Um, and obviously the code's been getting better over time. And so I, I feel like every time that I have this kind of like sense that, like, I'm special in, in some way, I'm often proven wrong. [laughs]

Ben Gilbert 28:18 Hmm.

Boris Cherny 28:19 And, and I think right now kind of with the current model capabilities, something that people talk a lot about is, like, the alpha is product taste, and I think this is also gonna go away.

David Rosenthal 28:28 Hmm.

Boris Cherny 28:28 Like, this is the alpha today, but also, you know, right now I have, you know, like, a couple hundred clouds running doing stuff, and a bunch of them are looking at, like, Twitter feedback, and they're looking at, like, GitHub issues, and they're looking at Slack, and they're figuring out what to build next. And right now most ideas are bad, but maybe 20% are good. If you wait for the next model and, you know, if you project out maybe three months, six months, most ideas will probably be good. Um, and you know, there's gonna be some new, there's gonna be some new thing. There's gonna be something else that we're really grayed out, but I think in the long term this kind of erodes.

Ben Gilbert 28:57 A- a- any guesses at what it is? I- i- is there some final thing where humans are uniquely good at it?

Boris Cherny 29:09 I th- I think the final thing that we're gonna be teaching the model is values, and s- how to be, you know... The, the way that we teach our kids how to be good people, we're gonna teach the model how to be a good model.

Ben Gilbert 29:23 Fascinating.

David Rosenthal 29:23 Fascinating.

Ben Gilbert 29:24 Well, that-

David Rosenthal 29:25 Yeah

Ben Gilbert 29:25 ... is quite the place to leave it. [laughs]

David Rosenthal 29:27 Yeah. Boris, thank you so much. This was fascinating.

Ben Gilbert 29:31 Thank you. [applause] [outro music]