you2idea@video:~$ watch PQU9o_5rHC4 [50:10]
// transcript — 790 segments
0:01 At Enthropic, the way that we thought about it is we don't build for the model
0:05 of today. We build for the model six months from now. That's actually like
0:08 still my advice to to founders that are building on LLM. Just try to think about
0:12 like what is that frontier where the model is not very good at today cuz it's
0:15 going to get good at it. All of Quad Code has just been written and rewritten
0:18 and rewritten and rewritten over and over and over. There is no part of Quad
0:21 Code that was around 6 months ago. You try a thing, you give it to users, you
0:24 talk to users, you learn, and then eventually you might end up at a good
0:27 idea. Sometimes you don't. Are you also in the back of your mind thinking that
0:30 maybe like in 6 months you won't need to prompt that explicitly? Like the model
0:33 will just be good enough to figure out on its own? >> Maybe in a month,
0:38 >> no more need for plan mode in a month. Welcome to another episode of the light
0:51 cone and today we have an extremely special guest, Boris Churnney, the
0:57 creator engineer of Claude Code. Boris, thanks for joining us.
0:59 >> Thanks for having me. >> Thanks for creating a thing that has
1:02 taken away my sleep for about 3 weeks straight. >> I am very addicted to Cloud Code and uh
1:10 it feels like rocket boosters. Has it felt like this for people like for you
1:13 know months at this point. I think it was like end of November is where uh a
1:18 lot of my friends said like something changed. >> I remember for me I felt this way when I
1:23 first created Quad Code and I didn't yet know if I was on to something. I kind of
1:26 felt like I was on to something and then that's when I wasn't sleeping.
1:28 >> Yeah. >> And that was just like three straight
1:30 months. >> This was uh September 2024. Yeah. It was like three straight months. I I didn't
1:36 take a single day vacation. Worked through the weekends. Worked every
1:39 single night. I was just like, "Oh my god, this is I think this is going to be
1:43 a thing. I don't know if it's useful yet because it it couldn't actually code
1:45 yet." >> If you look back on uh those moments to now, like what would be like the most
1:52 surprising thing about this moment right now? >> It's unbelievable that we're still using
1:55 a terminal. That was supposed to be the starting point. I didn't think that
1:58 would be the ending point. And then the second one is that it's even useful cuz
2:02 uh you know at the beginning it didn't really write code. Even in February when
2:06 we G it wrote maybe like 10% of my code or something like that. I didn't really
2:08 use it to write code. it wasn't very good at it. I still wrote most of my
2:13 code by hand. Uh so the fact that it it actually like our bets paid off and it
2:18 got good at the thing that we thought it was going to get good at because it
2:21 wasn't obvious. At Enthropic, the way that we thought about it is we don't
2:24 build for the model of today. We build for the model 6 months from now. And
2:28 that's actually like still my advice to to founders that are building on LLM is,
2:32 you know, just try to think about like what is that frontier where the model is
2:36 not very good at today. um because it's going to get good at it and you just
2:38 have to wait. >> Going back though, but when do you remember when you first got the idea?
2:42 Can you just talk us through that? Like was it some like a spark or what was
2:46 even the first version of it in your mind? >> You know, it's funny. It was like it was
2:49 so accidental that it just kind of evolved into this. Um you know as as
2:54 anthropic I think for Ant the bet has been coding for a long time and the bet
3:01 has been the path to save to safe AGI is through coding >> and this is this has kind of always been
3:06 the idea and the way you get there is you you teach the model how to code then
3:08 you teach it how to use tools then you teach it how to use computers um and you
3:12 can kind of see that because the the first team that I joined at Enthropic it
3:16 was called the anthropic labs team uh and it produced three products it was
3:21 quadcode MCP and in the desktop app. So you can kind of see how these like weave
3:24 together. The particular product that we built, you know, like no one no one
3:29 asked me to build a CLI. Um we kind of knew maybe it was time to
3:33 build some kind of coding product cuz it seemed like the model was ready, but no
3:37 one had yet really built the product that harnessed this capability. So like
3:41 still there's this insane feeling of product overhang. But at the time it was
3:44 just like even crazier cuz like no one had built this yet. And so I I started
3:49 like hacking around uh and I was like, "Okay, we build a coding product. What
3:52 do I have to do first? I have to understand how to use the API because I
3:55 hadn't used anthropic API at that point." Um and so I I just built like a
4:00 little terminal app to use the API. That's all that I did. And it was a
4:02 little chat app because you know like you think about the you know AI
4:05 applications of the time and you know for non-coders today most what what are
4:09 most people using is just a chat app. So that's what I built. Uh and you know it
4:13 was in a terminal. I can ask questions. I give answers. Then I think tool use
4:17 came out. I just wanted to try out tool use because I I don't really understand
4:20 what this is. I was like to use this is cool. Is this actually useful? Probably
4:23 not. Let me just try it. >> You built it in terminal just because it
4:27 was the easiest way to get something up and running. >> Yes. Cuz I didn't have to build a UI.
4:30 >> Okay. >> It was just me >> at that point. It was like the IDEs,
4:34 Cursor, Windsurf taking off. Were you sort of under any pressure or getting lots of suggestions
4:40 of, hey, like we should build this out as a plugin or as a as a fully featured
4:44 ID itself? There was no pressure because we didn't even know what we wanted to
4:47 build. Like the the team was just in explore mode, you know, like we we
4:50 didn't we know vaguely we wanted to do something in coding, but it wasn't
4:53 obvious what no one was high confidence enough. That was like my job to figure
4:57 out. And so I g I gave the model uh the batch tool. That was the first tool that
5:00 that I gave it just cuz I think that was literally the example in our docs. I
5:04 just like took the example. It was in Python. I just ported it to TypeScript
5:07 because that that's how I wrote it. You know, I didn't know like what the model
5:09 could do with bash. So I asked it to like read a file. It could like cat the
5:13 file. So like that was cool. And then I was like, "Okay, like what can you
5:16 actually do?" And and I asked her, "What music am I listening to?" He wrote some
5:21 like Apple script to script my my Mac and look up the music in my music
5:23 player. >> Oh my god. >> And this was Sauna 3.5.
5:28 >> And you know, like I I didn't think the model could do that. And that was my
5:33 first I think ever fuel the AGI moment >> where I was just like, "Oh my god, the
5:37 model it just wants to use tools. That's all it wants." >> That's kind of fascinating. I mean it's
5:44 very kind of contrarian that clocker works so well in such an elegant simple
5:49 form factor. I mean terminals have been around for a really long time and that
5:53 seemed to be like a good design constraint that allowed a lot of
5:58 interesting developer experiences like it doesn't feel like working. It just
6:02 feels fun as a developer. I don't think about files where everything is and that
6:07 came by accident almost. >> Yeah, it was an accident. I remember so
6:11 after the terminal started to take off internally. Um and honestly like after
6:15 building this thing I think like 2 days after the first prototype I started
6:18 giving it to my team just for dogfooting cuz you know like you know if you come
6:22 up with an idea and it seems useful the first thing you want to do is you want
6:24 to give it to people to see how they use it. And then I came in the next day and
6:28 then Robert who sits across from me who's another engineer he he just like
6:32 had quad code on his computer and he was like using it to code. I was like I was
6:35 like what what are you what are you doing? Like this thing isn't ready. It's
6:38 just a prototype. But yeah, it it was already useful in that form factor. And
6:42 I remember when we did our launch review to kind of launch quad code externally,
6:47 this was in December, November, something like that in 2024. Um Dario
6:52 asked and he was like, "The us chart internally like the the Dow chart is
6:55 like vertical. Are you like forcing engineers to use it? Like why are you
6:58 mandating them?" >> And I was just like, "No, no, we didn't.
7:02 We I just like posted about it and they they' just been like telling each other
7:05 about it." Honestly, it was it was just accidental. We we started with the CLI
7:08 because it was the cheapest thing and it just kind of stayed there for a bit.
7:13 >> So in that 2024 period, what how were the engineers using it? Were they sort
7:16 of shipping code with it yet or were they using it in a different way?
7:19 >> The model is not very good at coding yet. I I was using it personally for
7:24 automating git. Um I think at this point I I probably forgotten most of my git
7:28 because cloud code has just been doing it for so long. But yeah, like
7:31 automating uh bash commands that that was a very early use case and like
7:35 operating like Kubernetes and kind of things like this. People were using it
7:37 for coding. So there were some early signs of this. I think the first use
7:40 case was actually writing unit tests because it's a little bit lower risk and
7:44 the model was still pretty bad at it >> but people were were were kind of
7:47 figuring it out and and they were figuring out how to use this thing.
7:51 >> Um and one thing that we saw is people started writing these markdown files for
7:55 themselves and then having the model read that markdown file. And this is
7:59 where QuadMD came from. Probably the single for me biggest principle in
8:03 product is latent demand. Um and the just every bit of this product is built
8:06 through latent demand after their initial CLI. Uh and so quadmd is an
8:10 example of that. There's this other general principle that I think is maybe
8:15 interesting where you can build for the model and then you can build scaffolding
8:18 around the model in order to improve performance a little bit and depending
8:22 on the domain you can improve performance maybe 10 20% something like
8:26 that and then essentially the gain is wiped out with the next model. So either
8:30 you can build build the scaffolding and then you know get some performance gain
8:33 and then rebuild it again or you just wait for the next model and then you
8:36 kind of get it for free. the quantumd and kind of the scaffolding is an
8:39 example of that and really I think that's why we stayed in the CLI is
8:44 because we felt there is no UI we could build that would still be relevant in 6
8:48 months because the model was improving so quickly >> earlier we were saying like we should
8:52 compare cloud MDs but you said something very profound which is you know yours is
8:57 actually very short which is almost like the opposite of what you know people
9:01 might expect why is that what's in your cloud MD >> okay so I I checked this before we came
9:06 so my my cloud has two Um, one is, uh, there it's just two lines. So, the first line is whenever
9:14 you put up a PR, enable automerge. Um, so as soon as someone accepts it, it's
9:17 merged. That's just so I can like code and I don't have to kind of go back and
9:21 forth with CR or whatever. And then the second one is whenever I put up a PR,
9:24 post it in our internal team stamps channel. Uh, just so someone can stamp
9:29 it and I can get unblocked. Uh, and the idea is every other instruction is in
9:33 our quadmd that's checked into the codebase and it's something our entire
9:37 team contributes to multiple times a week. And very often I'll see someone's
9:43 PR and they make some like mistake that's totally preventable and I'll just
9:46 literally tag Claude on the PR. I'll just do like add quad, you know, like
9:49 add this to the quad MD and I'll do this, you know, like many times a week.
9:52 >> Do you have to like compact the Claude MD? Like I definitely reached a point
9:56 where I got the message at the top saying your cloud MD is like thousands
10:00 of tokens now. What do you do when you guys hit that? >> So our quadm is actually pretty short. I
10:05 think it's like couple thousand tokens maybe something like that. Um if if you
10:08 hit this my recommendation would be delete your quadmd and just start fresh.
10:11 >> Interesting. >> I think a lot of people like they try to
10:15 overengineer this right and and really like the capability changes with every
10:19 model. And so the thing that you want is do the minimal possible thing in order
10:23 to get the model on track. And so if you delete your quadd and then you know the
10:26 model is getting off track, it does the wrong thing. That's when you kind of add
10:30 back a little bit at a time. And what you're probably going to find is with
10:32 every model, you have to add less and less. For me, I consider myself a pretty
10:36 average engineer to be honest. Like I don't use a lot of fancy tools. Like I I
10:40 don't use like Vim. I use, you know, VS Code because it's simpler. Um I don't
10:42 really >> Wait, really? I would have assumed that because you built this in the terminal
10:47 that you were sort of like a dieh hard ter terminal like Vim Vim only person
10:51 you know screw those VS code people you know >> well we have people like that on the
10:54 team there's you know like Adam Wolf for example he's he's on the team he's like
10:58 you will never take Vim for my cold dead hands like yeah so there's definitely a
11:01 lot of people like that on the team and this is one of the things that I learned
11:04 early on is every engineer likes to hold their dev tools differently they like to
11:07 use different tools there's just no one tool that works for everyone but I think
11:10 also this is one of the things that makes it possible for quad code to be so
11:15 good because I kind of think about it as what is the product that I would use
11:19 that makes sense to me and so to use quad code you don't have to understand
11:22 Vim you don't have to understand TMX you don't have to know how to like SSH you
11:25 don't have to know all the stuff you just have to open up the tool and it'll
11:28 guide you it'll it'll do all this stuff >> how do you decide how verbose you want
11:32 like sort of the terminal to be like sometimes you have to go you know
11:36 control O and check it out and is it like internal bike shed battles around
11:42 like longer shorter I mean every user probably has a for an opinion like how
11:45 do you make those sorts of decisions? >> What What's your opinion? Is it is it
11:48 too verbose right now? >> Oh, I love the verbosity cuz basically
11:52 sometimes it just like goes off the deep end and I'm watching and then I can just
11:55 read very quickly and it's like, "Oh, no, no, it's not that." And then I
11:59 escape and then just stop it and then it just like stops an entire bug farm like
12:03 as it's happening. I mean, that's usually when I didn't do plan mode
12:05 properly. >> This is something that we probably change pretty often. Um, I remember
12:11 early on, this is maybe six months ago, I tried to get rid of bash output just
12:14 internally just to like summarize it because I was like these giant long bash
12:18 commands, I don't actually care. And then I gave it to anthropic employees
12:22 for a day and everyone just revolted. I want to see my dash because it it
12:25 actually is quite useful for, you know, like for something like git output,
12:29 maybe it's not useful, but if you're running, you know, like Kubernetes jobs
12:31 or something like this, you actually do want to see it. We recently hit the hid
12:36 the file reads and uh file searches. So you'll notice instead of saying, you
12:40 know, like read food.md said, you know, like read one file, search searched one
12:44 pattern. And this is something I think we could not have shipped six months ago
12:47 because the model just was not ready. It would have, you know, it still read the
12:50 wrong thing pretty often. As a user, you still had to be there and kind of catch
12:53 it and debug it. But nowadays, I just noticed it's on the right track almost
12:57 every time. And because it's using tools so much, it's actually a lot better just
13:01 to summarize it. Um, but then we shipped it. Uh, we dog fooded it for like a
13:04 month and then people on GitHub didn't like it. Uh so there was a big issue
13:07 where people like no like I want to see the details and that was really great
13:11 feedback. Um and so we added a new verbose mode and so that's just like in
13:15 slash config you can enable verbose mode and if you want to see all the file
13:18 outputs you can continue to do that and then I posted on the issue and people
13:22 still still didn't like it which is again awesome because like my favorite
13:25 thing in the world is just hearing people's feedback and hearing how they
13:28 actually want to use it. Um and so we just like iterated more and more and
13:30 more to get that really good and to make it the thing that people want. I'm
13:34 amazed like how much I enjoy uh fixing bugs now. And then all you have to do is
13:40 uh have really good logging and then even just say like hey check out that
13:44 you know this particular object it messed up in this way and it like
13:47 searches the log. It figures everything out. It can like go into your you can
13:51 make a production tunnel and it'll look at your production DB for you. It's like
13:55 this is insane. Bug fixing is just going to sentry copy markdown. You know pretty
13:59 soon it's just going to be straight MCP. It's like an autobug fixing like and
14:05 test making sort of uh what's the new uh term they call it like a making a
14:08 startup factory. Oh yeah. >> Right. There's like all these concepts
14:13 now of rather than having to review the code, you know, I'm I'm old school, so I
14:18 like the verbosity. I like to say, "Oh, well, you're doing this, but I want you
14:21 to do that." Right? But there's a totally different school of thought now
14:26 that says like anytime an a real human being has to look at code uh that's bad.
14:30 >> Yeah. Yeah. Yeah. >> Which is fascinating. >> I think like Dan Chipper talks about
14:35 this a lot as kind of when whenever you see the model make a mistake try to put
14:38 in the quadmd try to put it in like skills or something like that so it's
14:42 reusable. But I I think there's this meta point that I actually struggle with
14:46 a lot. And I people talk about like agents can do this, agents can do that,
14:49 but actually what agents can do, it changes with every single model. And so
14:52 sometimes there's a new person that joins the team and they actually use
14:56 quad code more than I would have used it. >> And I'm just constantly surprised by
15:01 this. Like for example, there was a we had like a memory leak and we were
15:04 trying to debug it. Um and by the way, like Jared Sumar has just been on this
15:07 crusade killing all the memory leaks and it's just been amazing. But before Jared
15:11 was on the team, I had to do this and there was this memory leak. I I was
15:14 trying to debug it. And so I I took a heap dump. I opened it in DevTools. I
15:17 was looking through the profile. Then I was looking through the code and I I was
15:20 just trying to figure this out. And then another engineer on the team, Chris, he
15:25 just like asked Quad Code. He was like, "Hey, I think there's a memory leak. Can
15:28 you like run this?" And then like try to figure it out. And Quad Code like took
15:30 the heap dump. It wrote a little tool for itself to like analyze the heap
15:34 dump. And then it found the leak faster than I did. And this is just something I
15:40 have to constantly relearn because my brain is still stuck somewhere six
15:44 months ago at times. >> So what would be some advice for
15:49 technical founders to really become maximalists at the latest model release?
15:54 It sounds like people off of fresh off of school or that don't have any
15:58 assumptions might be better suited than maybe sometimes engineers who have been
16:02 working at it for a long time. And how do the experts get better? I think for
16:07 yourself it's kind of beginner mindset and uh I don't know maybe just like
16:12 humility like I feel like engineers as a discipline we've learned to have very
16:16 strong opinions and senior engineers are kind of rewarded for this in my old job
16:19 at a big company when I hired like architects and this kind of a type of
16:23 engineer you look for people that have a lot of experience and really strong
16:26 opinions but it actually turns out a lot of this stuff just isn't relevant
16:29 anymore and a lot of these opinions should change because the model is
16:33 getting better um so I think actually the biggest skill is people that can
16:37 think scientifically and can just think from first principles.
16:40 >> How do you screen for that when you try to hire someone now for for your team?
16:44 >> I sometimes ask about what's an example of when you're wrong. It's a really good
16:47 one. You know, some of these like classic behavioral questions like not
16:50 even coding questions I think are quite useful because you can see if people can
16:53 recognize their mistake in hindsight, if they can claim credit for the mistake
16:58 and if they learn something from it. And I think a lot of these like very senior
17:01 people especially there there are some founder types like this but I think
17:04 founders in particular are actually quite good at it. Um but other people
17:09 sometimes will never really take uh they'll never take the blame for a
17:12 mistake. But I don't know like for me personally I'm wrong probably half the
17:16 time. Like half my ideas are bad and you just have to try stuff and you know you
17:20 try a thing you give it to users you talk to users you learn and then
17:23 eventually you might end up at a good idea. Sometimes you don't. And this is
17:27 the skill that I think in in the past was very important for founders, but now
17:31 I think it's very important for every engineer. >> Do you think um you would ever hire
17:37 someone based on the uh claude code transcript of uh them working with the
17:42 agent cuz we're actively doing that right now. We just added uh just as a
17:47 test like you can upload a transcript of you coding a feature with cloud code or
17:52 codeex or whatever it is. Personally, I think that like it's going to work. I
17:56 mean, you can figure out uh how someone thinks, like whether they're looking at
18:00 the logs or not, like can they correct the agent if it goes off off the rails?
18:04 Like, does do they use plan mode? You know, when they use plan mode, do they
18:08 make sure that there are tests or you know, all of these different things
18:10 that, >> you know, do they think about systems? Do they even understand systems? Like,
18:14 there's just so much that's sort of embedded in that that I imagine. I just
18:19 want like a spider uh a spiderweb graph, you know, like in those video games like
18:23 NBA 2K. It's like, oh, this person's really good at shooting or defense. It's
18:26 like you could imagine a spiderweb graph of like, you know, someone's claude code
18:30 skill level. >> Yeah. What would what would the skills
18:32 be? What would be those? >> I mean, I think it's like systems
18:37 testing must be like user behavior. I mean, there's got to be a design part
18:41 like product sense maybe also just like automating stuff. Mhm. My favorite thing
18:46 in CloudMD uh for me is I have a thing that says for every plan decide whether
18:51 it's overengineered, underengineered, or perfectly engineered and why.
18:54 >> I think this is something that we're trying to figure out, too, cuz I I think
18:57 uh when I look at engineers on the team that I think are the most effective,
19:00 there's essentially two, it's very biodal. Um there's one side where it's
19:05 extreme specialists. Um and so like I named Jared before, like he's a really
19:08 good example of this and kind of the bun team is a really good example. Just
19:11 hyper specialist. They understand dev tools better than anyone else. They
19:14 understand JavaScript runtime systems better than anyone else. And then
19:18 there's the flip side of kind of hyper generalists and that's kind of the rest
19:21 of the team. And a lot of people they span like product and info or product
19:25 and design um or you know like product and user research, product and business.
19:30 I really like to see people that just do weird stuff. I think that's one of these
19:33 things that was kind of a warning sign in the past because it's like can these
19:37 people actually build something useful? >> Um that's the limits test. Yeah, that's
19:42 what must but but nowadays like for example an engineer on the team Daisy,
19:45 she was on a different team and then she transferred onto our team and the reason
19:51 that I wanted her to transfer is she put up a PR for Claude Code like a couple
19:56 weeks after she joined or something and the PR was to add a new feature to
19:59 Claude Code and then instead of just adding the feature what she did is first
20:03 she put up a PR to give Claude code a tool so that it can test an arbitrary
20:07 tool and verify that that works. And then she put up that PR and then she had
20:11 Quad write its own tool instead of herself implementing it. And I think
20:14 it's this kind of out of the box thinking that is is just so interesting
20:17 because not a lot of people get it yet. You know, like we use the Quad agents
20:21 SDK to automate pretty much every part of development. It automates code
20:25 review, security review. Uh it labels all of our issues. It shephards things
20:29 to production. It does pretty much everything for us. But I think
20:32 externally I'm seeing a lot of people start to figure this out, but it's
20:35 actually taken a while to figure out how do you use LMS in this way? How do you
20:40 use this new kind of automation? So it's kind of a new skill.
20:42 >> I guess one of the uh funnier things that I've been having office hours with
20:46 various founders about is um you have like sort of the visionary founder who
20:50 has like the idea they've like built this like crystal palace of the product
20:55 that they want to build. they've totally loaded in their brain, you know, who the
20:59 user is and what they feel and what they're motivated by and then they're
21:03 sitting in claude code and they can do like, you know, 50x work and then but
21:07 they have engineers who work for them who like don't have the, you know,
21:11 crystal memory palace of like the platonic ideal of the product that the
21:15 pro founder has and they can only do like 5x work. Are you hearing stories
21:20 like that? there's usually a person who's like the core like designer of a
21:25 thing and they're just like, you know, trying to blast it out of their brain.
21:30 What's the nature of like teams like that? You know, it seems like that's
21:33 almost a stable configuration. Like you're going to have the visionary who
21:38 like now is unleashed, but you know, maybe going back to the top of it, like
21:41 I'm experiencing this right now. I was like, "Oh, well, I'm only a solo person
21:46 and you know, I need to eat and sleep and I have, you know, a whole job. It's
21:50 like, how am I going to do this?" You know, >> you know, like we just launched quad
21:54 teams and, you know, this is a way to do it, but you can also just build your own
21:57 way to do it. It's pretty easy. >> What's the vision for cloud teams?
22:01 >> Just collaboration. It's like there's this whole new field of like agent top
22:03 apologies that people are exploring. Like what are the ways that you can
22:07 configure agents? There's this one sub idea which is uncorrelated context
22:10 windows. And the idea is just multiple agents, they have fresh context windows
22:13 that aren't essentially polluted with each other's context or their own
22:17 previous context. And if you throw more context at a problem, that's like a form
22:21 of test time compute. Um, and so you just get more capability that way. And
22:24 then if you have the right topology on top of it, so the agents can communicate
22:27 in the right way, they're laid out in the right way, then they can just build
22:31 bigger stuff. And so Teams is kind of like one idea. There's a few more that
22:35 are coming pretty soon. Um, and the idea is just maybe it can build a little bit
22:40 more. I think the first kind of big example where it worked is our plugins
22:44 feature was entirely built by a swarm over over a weekend. It just ran for
22:47 like a few days. There wasn't really human intervention. And plugins is
22:51 pretty much in the form that it was when when it came out. >> How did you set that up? Like did you
22:56 spec out sort of the outcome that you were hoping for and then let it sort of
23:01 figure out the details and then like let it run? >> Yeah. an engineer on the team just gave
23:08 uh gave Quad a spec and um told Quad to use a Asauna board and then Quad just
23:13 put up a bunch of tickets on a sauna and then spawned a bunch of agents and the
23:16 agent started picking up tasks. The main quad just gave it instructions and they
23:19 all just figured it out >> like independent um agents that didn't
23:23 have the context of the bigger spec. Right. >> Right. If you if you think about the way
23:27 that uh you know like how our agents actually started nowadays and you know I
23:30 haven't pulled the data on this but I would bet the majority of agents are
23:34 actually prompted by quad today in the form of uh sub agents cuz like a sub
23:38 agent is just like a recursive quad code that's all it is in the code and it's
23:43 just prompted by we call her mama quad >> and that that's all it is and I think
23:46 probably if you look at most agents they're launched in this way
23:49 >> my claude insights just told me to do this more for debugging so that I get
23:53 like I spend a lot of time on debugging And it would just be better to have like
23:57 multiple sub agents spin up and like debug something in parallel. And so then
24:01 I just like added that to my claude MD to just be like, hey, like next time you
24:04 try and fix a bug like have one agent that like looks in the log, like one
24:07 that looks in the code path. That just seems sort of inevitable.
24:11 >> For weird scary bugs, I try to uh fix bugs in plan mode and then it seems to
24:15 use the agents to sort of search everything. Whereas like when you're
24:19 just trying to do it in line, it's like, okay, I'm going to do like this one task
24:23 instead of search wide. This is something I do all the time too. I I
24:26 just say if the if the test seems kind of hard, this kind of research test,
24:29 I'll calibrate the number of sub aents I ask it to use based on the difficulty of
24:32 the task. >> So if it's like really hard, I'll say like use three or maybe five or even 10
24:38 sub aents, research in parallel and then see what they come up with.
24:40 >> I'm curious. So then why don't you put that in your clawed MD file?
24:44 >> It's kind of case by case, you know, like quadm like what is it? It's just a
24:48 it's a shortcut. Like if you find yourself repeating the same thing over
24:52 and over, you put in the quad MD. But otherwise, you don't have to put
24:54 everything there. You can just prompt quad. >> Are you also in the back of your mind
24:57 thinking that maybe like in six months, you won't need to prompt that
25:01 explicitly? Like the model will just be good enough to figure out on its own.
25:05 >> Maybe in a month. >> No more need for plan mode in a month.
25:08 >> Oh my god. >> I think plan mode probably has a limited
25:10 lifespan. >> Interesting. >> That's some alpha for everyone here.
25:14 What would the world look like without plan mode? Do you just describe it at
25:18 the prompt level and it would just do it? One shot it? Yeah, we've uh we've
25:21 started experimenting with this because quad code can now enter plan mode by
25:25 itself. I don't know if you've you guys have seen that. >> Yeah.
25:28 >> So, we're trying to kind of get this experience really good. So, it would
25:32 enter plan mode the same point where a human would have wanted to enter it. So,
25:35 I think it's like I think it's something like this, but actually plan mode
25:38 there's no there's no big secret to it. All it does is it adds one sentence to
25:42 the prompt that's like please don't code. >> That's all it is. You can you can
25:45 actually just say that. >> Yeah. So it sounds like a lot of the
25:48 feature development for clock code is very much a what we talk about a YC talk
25:52 to your users >> and then you come and implemented it. It
25:55 wasn't the other way that you had this master plan and then implemented all the
25:58 features. >> Yeah. Yeah. I mean that that's all it was like plan mode was we saw users that
26:03 that were like hey quad come up with an idea plan this out but don't write any
26:06 code yet. And there was kind of various versions of this. Sometimes it was just
26:09 talking through an idea. Sometimes it was these very sophisticated specs that
26:13 that they were asking Claude to write, but the common dimension was do a thing
26:17 without coding yet. And so literally like this was like Sunday night at 10
26:21 p.m. I was I was just like looking at GitHub issues and kind of seeing what
26:24 people were talking about and looking at our internal Slack feedback channel and
26:27 I just wrote this thing in like 30 minutes and then uh shipped it that
26:31 night. It went out Monday morning. That was plan mode. So do you mean that there
26:34 will be no need for plan mode to in the sense of I'm worried that the model's
26:38 going to do like it's going to do like the wrong thing or head off in the wrong
26:41 direction but there will still be a need for that. You need to think through the
26:45 idea and figure out exactly what it is that you want and you have to do that
26:47 somewhere. >> I kind of think about it in terms of like kind of increasing model
26:53 capabilities. So maybe 6 months ago a plan was insufficient. So you get Claude
26:57 to make a plan. Let's say even with plan mode you still have to kind of sit there
27:00 and babysit cuz it can go off track. Nowadays what I do is probably 80% of my
27:03 sessions I say I say plan mode has a limited lifespan but I I'm a heavy plan
27:08 mode user. Um I probably 80% of my sessions I start in plan mode and claude
27:13 will you know it'll start it'll start making a plan. I'll move on to my second
27:16 terminal tab and then I'll have it make another plan and then when I run out of
27:20 tabs I open the desktop app and then I go to the code tab and then I just start
27:22 a bunch of tabs there and they all start in plan mode probably know like 80% of
27:26 the time. Once the plan is good, and sometimes it takes a little back and
27:30 forth, they just get clawed to execute. And nowadays, what I find with Opus 4.5,
27:35 I think it started with 4.6 it got really good. Once the plan is good, it
27:39 just stays on track and it'll just do the thing exactly right almost every
27:42 time. And so, you know, before you had to babysit after the plan and before the
27:47 plan, now it's just before the plan. So, maybe the next thing is you just won't
27:49 have to babysit. You can just kind of give a prompt and Quad will figure it
27:51 out. >> The next step is Claude just speaks to your users directly.
27:58 Yeah, it just bypasses you entirely. >> It's funny. This is actually the current
28:01 stuff for us. Our quads actually like they talk to each other. They talk to
28:05 our users on Slack, at least internally pretty often. Um, my quad will like
28:08 tweet once in a while. >> No way. >> Um, but I actually like delete it. It's
28:13 just like it's a little like cheesy. Like I don't love the tone.
28:16 >> What does it want to tweet about? >> Sometimes it'll just like respond to
28:19 someone cuz I always have like co-work in the background and it's like it's the
28:22 co-work that really loves to do that because it likes using a browser.
28:25 >> That's funny. A a really common pattern is I ask Quad to build something. It'll
28:29 look in the codebase. Uh it'll see some engineer touch something in the git
28:32 flame and then it'll message that engineer on Slack. Um just like asking a
28:36 clarifying question and then once it gets answer back, it'll keep going.
28:40 >> What are some tips for founders now on how to build for the future? Sounds like
28:44 everything is really changing. What are like some principles that will stay on
28:48 and what will change? >> So I think some of these are pretty are
28:51 pretty basic, but I think they're even more important now than they were
28:55 before. Um, so one example is latent demand. Like I mentioned it a thousand
28:58 times for me. It's just like the single biggest idea in product. It's a it's a
29:03 thing that no one understands. It's a thing I certainly did not understand my
29:07 first few startups. And and the idea is like people will only do a thing that
29:10 they already do. You can't get people to do a new thing. If people are trying to
29:13 do a thing and you make it easier, that's a good idea. But if if people are
29:16 doing a thing and you try to make them do a different thing, they're not going
29:19 to do that. And so you just have to make the thing that they're trying to do
29:22 easier. And I think quad is going to get increasingly good at kind of figuring
29:25 out these kind of product ideas for you just because it can look at feedback, it
29:28 can look at debug logs, it can kind of figure this out. >> That's what you mean by plan mode was
29:32 latent demand that people were already like I don't know had their clawed chat
29:36 window open in a browser and were like talking to it to figure out like the
29:42 spec and and what it should do. And now that like pi mode just became that you
29:44 just do it in claw code. >> Yeah. Yeah, that's it. Some sometimes
29:47 what I'll do is I'll just walk around the office on on our floor and I'll just
29:52 kind of stand behind people like I I'll say like hi so it's not and then um I'll
29:55 I'll just see kind of like how they're using quad code. Um and this is also
29:59 just something I saw a lot um but it also came up in GitHub issues like
30:02 people were talking about it. It seems like so you're surprised how far the
30:06 terminal has gone and how far it's been pushed like how far do you think it has
30:11 left to go just given with this world of swore multiple agents like do you think
30:16 there's going to be a new a need for a different UI on top of it?
30:19 >> It's funny if you asked me this a year ago I would have said the terminal has
30:21 like a threemonth lifespan and then we're going to move on to the next
30:24 thing. Uh and you can see us experimenting with this right because
30:27 quad code started in a terminal but now it's in you know it's on web you can
30:32 like quadcode it's in the desktop app you know we've had that for you know
30:34 like three months or six months or something just in the code tab um it's
30:39 in the iOS and Android apps just like in the code tab it's in slack it's in
30:43 GitHub there's VS Code extensions there's Jet Brains extensions so we're
30:46 just like we're always experimenting with different form factors for this
30:50 thing to figure out what's the next thing I've been wrong so far about the
30:54 of the CLI. So, I'm probably not the person to forecast that.
30:58 >> What about like your advice to DevTool founders? Like, someone's building a
31:02 DevTool company today. Should they just like be building for engineers and
31:06 humans or should they be thinking more about like what Claude going to think
31:11 and want and build for sort of like the agent? >> The way I would frame it is think about
31:16 the thing that the model wants to do and figure out how do you make that easier.
31:20 And that's something that we saw, you know, like when I first started hacking
31:23 on quad code, I I realized like this thing just wants to use tools. It just
31:27 wants to interact with the world. And how how do you how do you enable that?
31:31 Well, the way you don't do it is you put it in a box and you're like, here's the
31:34 API, here's how you interact with me, and here's how you interact with the
31:37 world. The way you do it is you see what tools it wants to use. You see what it's
31:41 trying to do, and you enable that the same way that you do for your users. And
31:44 so, like for if you're building a dev tool startup, I would think about like
31:47 what is the problem you want to solve for the user? And then when you use when
31:50 you apply the model to solving this problem, what is the thing the model
31:52 wants to do? >> And then what is the technical and product solution that serves the weight
31:58 and demand of both? YC's next batch is now taking applications. Got a startup
32:04 in you? Apply at y combinator.com/apply. It's never too early and filling out the
32:09 app will level up your idea. Okay, back to the video. Back in the day, more than
32:15 10 years ago, you were a very heav heavy user and you wrote a book about
32:20 TypeScript, right? Before Typescript was cool. This is when everyone was a deep
32:24 in JavaScript. This is back in early 2010s, right? >> Yeah, something like that.
32:29 >> Before Typescript was a thing because back then is a very weird language. It's
32:34 not supposed to do a lot of things with being typed in JavaScript and now it's
32:39 the right thing and it feels like clot code in the terminal has a lot of
32:43 parallels with TypeScript at the beginning. >> TypeScript makes a lot of really weird
32:49 language decisions. So if you look at the type system pretty much anything can
32:53 be a literal type for example and this is like this is super weird cuz like
32:57 even like like Haskell doesn't even do this. It's just like it's too extreme or
33:02 it has like conditional types which I don't think any language thought of at
33:05 all. >> It was like very strongly typed. >> Yeah, it was very strongly and and the
33:10 idea was like when you know like when Joe Pamer and Anders and the early team
33:14 was like building this thing, the way they built it is we okay, we have these
33:17 teams with these big untyped JavaScript code bases. We have to get types in
33:20 there, but we're not going to get engineers to change that the way that
33:23 they code. You're not going to get JavaScript people to have like, you
33:27 know, 15 layers of class inheritance like you would a Java programmer, right?
33:29 They're going to write code the way they're going to write it. They're
33:31 they're going to use reflection and they're going to use mutation and
33:34 they're going to use all these features that traditionally are very very
33:36 difficult to type. >> They're a very unsafe type to any strong
33:40 functional programmer. >> That's right. That's right. That's
33:42 right. And so the thing that they did instead of getting people to kind of
33:45 change the way that they code, they they built a type system around this. And it
33:50 was just it's brilliant because there's all these ideas that no one was thinking
33:53 about even in academia like no one thought of a bunch of these ideas. It
33:57 purely came out of the practice of observing people and seeing how
33:59 JavaScript programmers want to write code. And so you know for for quad code
34:04 it there there are some ideas that are kind of similar in that you know like
34:07 you can use it like a Unix utility. You can pipe into it. You can pipe out of
34:10 it. Um in some ways it is kind of rigorous in this way but in in almost
34:15 every other way it's just the tool that we wanted. like I I build a tool for
34:18 myself and then the team builds the tool for themselves and then for anthropic
34:23 employees and then for users and it just ends up being really useful. It's not
34:26 it's not this like principled and academic thing which I think the the
34:31 proof is actually in the results. Now fast forward more than 15 years later
34:36 not many codebases are in Haskell which is more academic and there's tons of
34:40 them now on TypeScript because it's way more practical >> right
34:43 >> which is interesting. Yeah, it is interesting, right? It's like TypeScript
34:45 solves a problem. >> I guess one thing that's cool, I don't
34:48 know how many people know, but the terminal is actually one of the most
34:54 beautiful terminal apps out there and is actually written with React terminal.
34:58 >> When I first started building it, you know, like I I did front-end engineering
35:02 for for a while. So, and I was also like a, you know, I'm I'm sort of like a
35:06 hybrid, like I do like design and user research and, you know, write code and
35:09 all this stuff. And we love hiring engineers that are like this. Um, so we
35:13 just we love generalists. So for me it's like okay, I'm building a thing for the
35:15 terminal. I'm actually kind of a shitty Vim user. So like how do I build a thing
35:20 for people like me that um you know are are going to be working in a terminal.
35:24 And I think just the delight is so important. And I feel like at YC this is
35:27 something you talk about a lot, right? It's like build a thing that people
35:30 love. If the product is useful but you don't fall in love with it, that's not
35:34 great. Um so it kind of has to do both. Designing for the terminal honestly has
35:37 been hard, right? It's like uh it's like 80 by 100 characters or whatever. you
35:41 have like 256 colors, you have one font size, you don't have like mouse
35:45 interactions, there's all this stuff you can't do, and there's all these very
35:48 hard trade-offs. So, like a little known thing, for example, is you can actually
35:51 enable mouse interactions in a terminal. So, you can enable like clicking and
35:53 stuff. >> Oh, how do you do that in cloud code? I've been trying to figure out how to do
35:56 this. >> We don't we don't have it in cloud code because we actually prototyped it a few
36:00 times and it felt really bad because the trade-off is you have to virtualize
36:04 scrolling and so there's all these weird trade-offs because like the way
36:07 terminals work is like there's no DOM, right? It's like there's like anti-
36:11 escape codes and these kind of weird organically evolved specs since like the
36:14 1960s or whatever. >> Yeah. It feels like BBS's. It's like a
36:17 BBS door game. >> Yeah. >> Oh my god. >> That's like that's like a great
36:21 compliment. Yeah. Yeah. Like it should feel like you're discovering
36:24 >> Lord of the Red Dragon. It's fantastic. Oh my god. >> Yeah. But we have we've had to just like
36:28 discover all these kind of UX principles for building the terminal cuz no one
36:32 really writes about this stuff. And if you look at the big terminal apps of,
36:35 you know, like the 80s or 90s or 2000s or whatever, they use like ed curses and
36:39 they have all these like windows and things like this. And it just looks kind
36:42 of like janky by modern standards. It just looks too heavy and complicated.
36:46 And so we had to like reinvent a lot. And you know, for example, something
36:49 like the terminal spinner, like just like the spinner words, it's gone
36:54 through probably I want to say like 50 maybe 100 iterations at this point. And
36:58 probably 80% of those didn't ship. So we tried it, it didn't feel good, move on
37:01 to the next one. try it, didn't feel good, move on to the next one. Uh, and
37:04 this was like sort of one of the amazing things about quad code, right? Is like
37:07 you can write these prototypes and you can just do like 20 prototypes back to
37:11 back, see which one you like, and then ship that and the whole thing takes
37:13 maybe a couple hours. >> Whereas in the past, what you would have
37:16 had to do is like wen to use origami or framer or something like this. You built
37:19 like maybe three prototypes, it took like two weeks. It just took much much
37:22 longer. >> And so we have this luxury of we have to discover this new thing. We have to
37:26 build a thing. We don't know what the right endpoint is, but we can iterate
37:31 there so quickly and that's what makes it really easy and that's what lets us
37:35 build a product that's like joyous and that people like to use.
37:38 >> Boris, you had other advice for for builders and we kept interrupting you
37:43 because we have so many questions, but >> I would say um so okay, so maybe two
37:48 pieces of advice that are kind of weird because it's like about building for the
37:52 model. So one is uh don't build for the model of today, build for the model of 6
37:55 months from now. This is like sort of weird, right? Because like you can't
37:58 find PMF if the product doesn't work. But actually this is the thing that you
38:00 should do because otherwise what will happen is you spend a bunch of work you
38:04 find PMF for the product right now and then you're just going to get leaprogged
38:07 by someone else um because they're building for the next model and a new
38:10 model comes out every few months. Use the model, feel out the boundary of what
38:13 it can do and then build for the model that you think will be the model maybe 6
38:16 months from now. I think the second thing is um you know actually in the in
38:20 the quad code where in the quad code area where we sit we have a framed copy
38:24 of the bitter lesson on the wall. Um and this is this like rich sutton uh I like
38:28 everyone should read it if if you haven't uh and the idea is the more
38:33 general model will always be the more specific model and there's a lot of
38:37 corlaries to this but essentially what it boils down to is never bet against
38:41 the model. Uh, and so this is just like a thing to that that we always think
38:45 about where we could build a feature into cloud code. We could make it better
38:48 as a product and we call this scaffolding. That's all this code that's
38:51 not the model itself. But we could also just wait like a couple months and the
38:55 model can probably just do the thing instead. Um, and there's always this
38:58 trade-off, right? It's like engineering work now and you can kind of extend the
39:02 capability a little bit, maybe 10 20% or whatever in whatever domain on this
39:05 like, you know, like the spider chart of what you're trying to extend. Um, or you
39:09 can just wait and the next model will do it. So just always always think in terms
39:12 of this trade-off where where do you actually want to invest and assume that
39:15 whatever the scaffolding is it's just tech. >> How often do you rewrite the code ways
39:21 of uh clock code is every six months with this with this >> is there scaffolding that you've deleted
39:25 because you don't need it anymore because the model just improved.
39:29 >> Oh so much. Yeah. Like all of quad code has just been written and rewritten and
39:31 rewritten and rewritten over and over and over. We unhip tools every couple
39:36 weeks. We add new tools every couple weeks. There's no part of quad code that was
39:40 around six months ago. It's just constantly rewritten. >> Would you say most of the code base for
39:46 current cloud code is only say 80% of it is only less than a couple months old.
39:50 >> Yeah, definitely. It might it might even be like less than Yeah, maybe like a
39:53 couple months. That that feels about right. >> So it's like the life cycle of code now.
39:57 That's another alpha is expecting it to be the shelf life to be just couple
39:58 months. >> Yeah. >> For the best founders. >> Do you see uh Steve Yaggi's uh post
40:05 about how awesome working at Anthropic is? And I think there's a line in there
40:09 that says that an anthropic engineer uh currently averages 1,000x more
40:15 productivity than a Google engineer at Google's peak which is really an insane
40:22 number honestly like 1,000x like you know we're 3 years ago we were still
40:25 talking about 10x engineers now we're talking about 1000x on top of a Google
40:29 engineer in the prime like this is unbelievable honestly. Yeah, I mean
40:32 internally if you if you look at like technical employees, they all use quad
40:36 code every day. Um, and even non-technical employees, I think like
40:39 half the sales team uses quad code. Um, they they've started switching to
40:42 co-work because it's a little easier to use. It has like a VM, so it's a little
40:45 bit safer. But yeah, we actually we just pulled a stat and the I think the team
40:50 doubled in size last year, but productivity per engineer grew something
40:53 like 70%. >> It's measured by >> just like the simplest stupidest
40:57 measure, pull requests. Um, but we also kind of cross check that against like
41:01 commits and like uh the lifetime of commits and things like this. And since
41:04 quad code came out, productivity per engineer at anthropic has grown 150%.
41:07 >> Oh my god. >> Um, and this is crazy because I in my
41:12 old life I was responsible for code quality at Meta. >> Um, and I was responsible for the
41:16 quality of all of our code bases across every product across like you know
41:20 Facebook, Instagram, WhatsApp, whatever. >> And one of the things that the team
41:24 worked on was improving productivity. And back then seeing a gain of something
41:28 like 2% in productivity that was like a year of work by hundreds of people. And
41:33 so this like 100% this is just like unheard of just completely unheard of.
41:36 >> What drove you to come over to Anthropic? I mean basically as a builder
41:40 you could go anywhere. What was the moment that made you say like actually
41:44 this is the set of people or this is the approach. I was living in rural Japan
41:50 and I was opening up Hacker News every morning and I was reading the news and
41:54 uh it was all it just started to be like AI stuff at some point and uh I started
41:59 to use some of these early products and uh I remember like the first couple
42:03 times that I used it I was just like it just took my breath away. That was like
42:05 very cheesy to say, but that was actually that was actually the feeling.
42:08 Like it was just like it was amazing like as a as as a builder, I've just
42:12 never kind of felt felt this feeling like using these very very early
42:15 products. That was like in the quad 2 days or you know something like that.
42:19 And so I I just talking started talking to friends at Labs um just to kind of
42:25 see what was going on. Um and uh I met Ben man who's one of the founders at uh
42:30 at Anthropic and uh he just immediately won me over. Um and as soon as I met
42:33 kind of the rest of the team at an it just won me over and I think I think
42:37 probably in two ways. So one is it operates as a research lab. Um so the
42:42 product was teeny teeny tiny. It's really all about building a safe model.
42:46 That's all that matters. Um and so this idea of just being very close to the
42:48 model and being very close to development and being not the most
42:52 important thing because the product isn't anymore. It's just the model is
42:56 the thing that's the most important. Um that really resonated with me after
42:59 building product for many years. And then the second thing was just how
43:04 missiondriven it is. Um like I'm I'm a huge sci-fi reader. My bookshelf is just
43:08 like filled with sci-fi. And so like I just know how bad this can go.
43:11 >> And when I kind of think about what's going to happen this year, it you know
43:15 it's going to be totally insane. And in the worst case it can go very very bad.
43:19 >> Um and so I just wanted to be at a place that really understood that and kind of
43:23 really internalized that. And at Ant, you know, like if you overhear
43:26 conversations in the lunchroom or in the hallway, people are talking about AI
43:30 safety. this is really the thing that everyone cares about more than anything.
43:33 Um, and so I just wanted to be in a place like that. I I know I know for me
43:36 personally the mission is just so important. >> What is gonna happen this year?
43:42 >> Okay. So if you think back like six months ago and uh kind of what are the
43:45 predictions that people are making? So Daario predicted that 90% of the code at
43:50 Anthropic would be would be written by Quad. This is true. Um for me personally
43:56 it's been 100% for like since Opus 4.5. Um I just I uninstalled my IDE. I don't
44:01 edit a single line of code by hand. It's just 100% quad code and Opus. Um and you
44:05 know I land you know like 20 PR a day every day. If you look at Enthropic
44:11 overall it ranges between like 70 to 90% uh you know depending on the team. For a
44:14 lot of teams it's also like 100% for a lot of people it's 100%. And I remember
44:18 making this prediction back in May when we ged cloud code that you wouldn't need
44:24 an ID to code anymore. Uh and it was totally crazy to say. I feel like people
44:27 in the audience gasped >> because it was such like a silly
44:29 prediction at the time. But really all it is is like you just like trace the
44:32 you know the exponential >> and this is just like so deep in you
44:37 know the DNA at cuz like you know three of our founders were co-authors of the
44:40 scaling laws paper they kind of they saw this very early and so this is just like
44:43 tracing the exponential this is what's going to happen and yes that happened.
44:47 So continuing to trace the exponential I think what will happen is coding will be
44:52 generally solved for everyone. Um, and I think today coding is practically
44:54 solved, you know, for me and I think it'll be the case for everyone. Um, you
44:58 know, regardless of domain, I think we're going to start to see the title
45:01 software engineer go away. And I think it's just going to be maybe builder,
45:05 maybe product manager, maybe we'll keep the title as kind of a vestigial thing,
45:09 but the work that people do, it's not just going to be coding. It's software
45:12 engineers are also going to be writing specs. They're going to be talking to
45:15 users. like this thing that we're starting to see right now in our team
45:18 where engineers are very much generalists and every single function on
45:23 our team codes like our PM's code, our designers code, our EM codes, our um
45:28 like everyone our our finance guy codes like everyone on our team codes. We're
45:31 going to start to see this everywhere. So this is sort of uh this is kind of
45:35 like the lower bound if we just continue the trend. The upper bound I think is a
45:40 lot scarier. Um, and this is something like, you know, we hit ASL4. Um, and
45:43 this, you know, at anthropic, we talked about these safety levels. ASL3 is where
45:48 the models are right now. ASL4 is the model is recursively self-improving. Um,
45:51 and so if this happens, essentially, we have to meet a bunch of criteria before
45:55 we can release a model. And so the the extreme is that, you know, this happens
45:58 um or there's some kind of catastrophic misuse like people are using the model
46:03 to design bioiruses, design zero days, stuff like this. Um, and this is
46:06 something that we're really really actively working on so that doesn't
46:10 happen. I think uh it's just been honestly it's just been like so exciting
46:13 and humbling like seeing how people are using quad code like uh you know I just
46:17 wanted to build a cool thing and it ended up being really useful uh and that
46:21 was so surprising and so exciting. >> My impression from Twitter or just the
46:25 outside is basically everyone went away over the holidays and then like found
46:28 out about Claude code and it's just been crazy ever since. Is that how it was for
46:33 you at like internally? Did you were you having like a nice Christmas break and
46:37 then came back and like what happened? Well, actually for all of December, I
46:40 was traveling around. Uh, and I I took a coding vacation. So, we were kind of
46:43 traveling around and I was just like coding every day. So, that was really
46:46 nice. Uh, and then I also started to use Twitter at the time cuz like I I worked
46:50 on Threads back then way back when. So, I've been a Threads user for a while.
46:53 So, I just like tried to see kind of like other platforms where people are.
46:56 Yeah. I think for a lot of people they kind of discovered that was the moment
46:59 where they discovered Opus 4.5. I kind of already knew. >> Mhm.
47:03 >> Uh, and internally quad code's just been on this like exponential tear for many
47:07 many months now. So that just like it it became even more steep. That's what we
47:12 saw. And if you look at cloud code now, you know, there was some stat from
47:15 Mercury that like 70% of startups are you know choosing cloud as their model
47:19 of choice. There was some other stat from like semi analysis that 4% of all
47:23 public commits are made by cloud code. um like of all code written everywhere.
47:27 All the companies, you know, use squad code from like the biggest companies to
47:31 kind of, you know, smallest startups, you know, like it it wrote it it plotted
47:35 the course for Perseverance like for like the Mars rover. This is just like
47:38 this is the coolest thing for me. And we like we even printed posters cuz the
47:41 team was like, "Wow, this is just like so cool that NASA chooses to use this
47:44 thing." So, yeah, it's just like it's humbling. Um but it also feels like the
47:48 very beginning. What's the sort of interaction between uh claude code and
47:53 then co-work like you know was it a fork of cla code? Was it like you had cla
47:59 code look at the cloud code and say let's make a new spec for nontechnical
48:03 people that you know keeps all the lessons and then you know it sort of
48:07 went off for a couple days and did that. What's the genesis of that and you know
48:11 where do you think that goes? >> This is going to be like my fifth time
48:15 using the word wait and demand. It was just that I mean like we we were
48:17 looking at Twitter and there was like that one guy that was using quad code to
48:20 like monitor his tomato plants. >> Mhm. >> Uh there was like this other person that
48:24 was using it to like recover wedding photos off of a corrupted hard drive.
48:28 There were people that using it for like uh for finance. When we looked
48:32 internally at anthropic, every designer is using it all the entire finance team
48:36 at this point is using it. The entire data science team is using it not for
48:39 coding. People are jumping over hoops to install a thing in the terminal so that
48:42 they could use this. So we knew for a while that we wanted to build something
48:45 and so we're experimenting with a bunch of different ideas and the thing that
48:49 kind of took off was just you know a little cloud code wrapper in a guey in
48:52 the desktop app and that's all it is. It's just quad code under the hood. It's
48:54 the same agent. >> Oh wow. >> Um and uh Felix and the team and Felix
49:00 was early Electron contributor. He kind of knows that stack really well and he
49:04 was hacking on various ideas and uh they they built it in I think something like
49:08 10 days. It was it was just like 100% written by quad code. Uh and it just
49:12 felt ready to release. There was a lot of stuff that we had to build for
49:15 nontechnical users. So it's a little bit different than a technical audience. Uh
49:20 it runs in a all the code runs in a virtual machine. Uh there's a lot of
49:24 delete uh protections for deletion and things like this. There's a lot of
49:27 permission prompting and kind of other guardrails for users. Um yeah, it was
49:33 honestly pretty obvious. Boris, thank you so much for making something that uh
49:38 is taking away all my sleep, but in return, it's making me feel creator mode
49:43 again, sort of founder mode again. It's been an exhilarating 3 weeks. I like
49:46 can't believe I waited that long since November to actually get into it. Thank
49:50 you so much for being with us. Thank you for building what you're building.
49:54 >> Yeah, thanks for having me. And uh send bugs. >> Sounds good.
$

Inside Claude Code With Its Creator Boris Cherny

@ycombinator 50:10 22 chapters
[AI agents and automation][content creation and YouTube][developer tools and coding]
// chapters
// description

A very special guest on this episode of the Lightcone! Boris Cherny, the creator of Claude Code, sits down to share the incredible journey of developing one of the most transformative coding tools of the AI era. 00:00 Intro 01:45 The most surprising moment in the rise of Claude Code 02:38 How Boris came up with the idea for Claude Code 05:38 The elegant simplicity of terminals 07:09 The first use cases 09:00 What’s in Boris’ CLAUDE.md? 11:29 How do you decide the terminal’s verbosity? 15:44 B

now: 0:00
// tags
[AI agents and automation][content creation and YouTube][developer tools and coding]