00:02:04 · Aditya Agarwal, Thariq Shihipar
[00:02:04] S1: Yeah. [00:02:04] S0: Yeah. I was just sort of going through it. You know, was sort of mourning my old company and trying to figure out like what, like, could I make another company or like, was I good at this AI stuff or something, you know? And talking to Tom who had like sort of led interpretability at DeepMind and sent me these papers, and I was like, oh, yeah. Like, I... It's not that hard. You know? Like, I can start like, you know, doing work here and contributing. And so I worked for Tom's company for about three Yeah. And so after SPC, I worked for Tom's company. Then while I was doing that and wrapping that up, Sawyer texted me being like, dude, have you used Claude Code? What month? This was like March 2025, I think. Yeah. And he's like, Opus four just came out. It's so much better than Cursor. And was like, no, I I I have it. But I I got into it and I was immediately just like really really hooked. And I was like, yeah, I need to like sort of join Anthropic and I like doesn't matter what they want me to do there, I'll I'll do it, you know? I think initially I joined as like a... What what do they call it? Like a a demo designer on the design team. And I was just like, dude, like if you think that's what I can do, like sure. You know what I mean? And then ultimately, like, very quickly found myself on the Claude Code team. Yeah.
00:02:04 · Thariq Shihipar, Aditya Agarwal
[00:02:04] S0: Yeah. I was just sort of going through it. You know, was sort of mourning my old company and trying to figure out like what, like, could I make another company or like, was I good at this AI stuff or something, you know? And talking to Tom who had like sort of led interpretability at DeepMind and sent me these papers, and I was like, oh, yeah. Like, I... It's not that hard. You know? Like, I can start like, you know, doing work here and contributing. And so I worked for Tom's company for about three Yeah. And so after SPC, I worked for Tom's company. Then while I was doing that and wrapping that up, Sawyer texted me being like, dude, have you used Claude Code? What month? This was like March 2025, I think. Yeah. And he's like, Opus four just came out. It's so much better than Cursor. And was like, no, I I I have it. But I I got into it and I was immediately just like really really hooked. And I was like, yeah, I need to like sort of join Anthropic and I like doesn't matter what they want me to do there, I'll I'll do it, you know? I think initially I joined as like a... What what do they call it? Like a a demo designer on the design team. And I was just like, dude, like if you think that's what I can do, like sure. You know what I mean? And then ultimately, like, very quickly found myself on the Claude Code team. Yeah. [00:03:21] S1: I guess on that note, you you have sort of become one of the faces of Claude Code. Was that intentional or organic? I mean, I think you've historically liked technical writing, but like, how did that come about?
00:03:21 · Aditya Agarwal, Thariq Shihipar
[00:03:21] S1: I guess on that note, you you have sort of become one of the faces of Claude Code. Was that intentional or organic? I mean, I think you've historically liked technical writing, but like, how did that come about? [00:03:32] S0: Yeah. I mean, I I think that my, like, specific interest was sort of like trying to increase the, like, communication bandwidth between humans and agents. And and so I feel that like the agents and these models overall are are much more intelligent than maybe even right now we're utilizing them. When I joined the Claude Code team, one of the first features I introduced was this thing called ask user questions. So like during the flow, especially during planning, Claude Code might be like, hey, let's stop and like clarify what you're trying to do. Right? And I I think previous to this, like, people would just sort of like... It was very dependent on the user to, in detail, explain what they want and know the unknown unknowns, for example. And so that was like something where I felt has become one of the most popular features in the Claude Code, really changed how people interact with the the agent. But I also felt the more time I've spent on it, I actually feel like it's more important to educate humans than improve the agent UX sometimes. Because as the models get smarter, there's just way more stuff that they can do. And like how do you build a mental model of how to prompt Claude is like really like maybe more valuable or like I'm better at it than like, you know, doing more product work.
00:04:50 · Aditya Agarwal
[00:04:50] S1: I'd be curious being on the inside and we'll get into like, we'll we'll unpack, you know, how to get the most out of Claude in a sec. But I think a lot of us had this moment in February of this year with 4.6. That was kind of this oh moment where the model got significantly better in the long horizon. Reasoning was at a point where you could actually see yourself doing, like, really meaningful big work. Internally, when was that moment for you guys or for you personally where you're like... Obviously, you you saw that March 2025 with four. But was there a moment in between in advance of when we saw publicly where it kinda hit you?
00:05:26 · Thariq Shihipar, Aditya Agarwal
[00:05:26] S0: Yeah. I mean, this is a good question. I I think that... Yeah. For me, the moment was Opus four. Not that Opus 4.5 was not better at doing long running tasks, but I just think you can sort of see the progression of it. And I think I was actually surprised by how popular Opus 4.5 was kind of. Like, I I thought it was a jump, but I I don't think I realized how much like, I think there are subtle things where like, if you're good enough at prompting, maybe the previous generation of models can do it. But like, then the next generation just does it automatically, you know? And so I think it it feels a lot like... Doesn't feel to me like a single eureka moment. I think it's like you spend time with the models all the time and you sort of develop these habits. And sometimes you find things that, like, you know, we've sort of, like, been exposed to for a while, and then we release to the world and and peep... Forget how big the jump can be sometimes. Yeah. [00:06:23] S1: So one last, like, quick hit intro. What was the first thought that came to mind when you heard that Karpathy was joining? [00:06:30] S0: And when was that? I mean, I'm excited. I don't do pre training, you know. So I I I think it's obvious... Obviously, like a good signal in the sense that, like, Karpathy, you know, cares about education a lot, but also is a great researcher. And, like, where do you spend your time, especially, like, automating pre training. Yeah. I mean, like, if he thinks that's the best place to spend his time, I think that's a strong signal.
00:06:30 · Thariq Shihipar, Aditya Agarwal
[00:06:30] S0: And when was that? I mean, I'm excited. I don't do pre training, you know. So I I I think it's obvious... Obviously, like a good signal in the sense that, like, Karpathy, you know, cares about education a lot, but also is a great researcher. And, like, where do you spend your time, especially, like, automating pre training. Yeah. I mean, like, if he thinks that's the best place to spend his time, I think that's a strong signal. [00:06:53] S1: Yeah. I mean, I think a lot of you have heard, like, a phrase we use here, but, you know, best founders being super talent greedy. And when I saw it, I was like, they're talent greedy. That's amazing. And magnetic to be able to land Andre there. That's awesome. Well, let's get kind of into the sort of little more of the the technical discussion. And something you've tweeted about, but also I know Claude has talked about is sort of, like, avoiding the capability overhang and, like, making sure that you're building sort of on the exponential. Let's spend a couple minutes. Like, why don't you expand on that? But I think what's most interesting maybe to this audience, especially given where you sit, is to the extent that you can sort of bring forward some live examples of of you guys building. It's very meta you guys are building Claude to then allow us to use Claude. But, like, in the ways that you build, how you actively think about those two concepts.