THARIQ SHIHIPAR (Anthropic) - Inside Claude Code

=== THARIQ SHIHIPAR (Anthropic) - Inside Claude Code ===

(06:02:20): Welcome.

(06:02:21): Hey, how's it going, guys?

(06:02:23): Can you hear me right?

(06:02:24): Yes.

(06:02:25): We can hear you.

(06:02:26): Where are you calling in from?

(06:02:28): Anthropic.

(06:02:29): So, yeah, yeah, yeah.

(06:02:32): Great.

(06:02:32): Possible, I don't know, maybe I get kicked out of this room at some point.

(06:02:36): We'll see.

(06:02:36): That's fine.

(06:02:37): That's fine.

(06:02:37): We would love a little virtual live stream tour of the office,

(06:02:41): so if you get kicked out,

(06:02:42): it's really cool.

(06:02:43): Um, so I mean, we're big, uh, obviously anthropic cloud code stands over here.

(06:02:50): So I'm, I'm like fanboying a little bit, just getting to talk to you.

(06:02:53): Thank you so much for,

(06:02:55): um,

(06:02:55): for being part of building cloud code and all this,

(06:02:57): all the stuff you're making.

(06:02:58): It's, it's been awesome.

(06:02:59): And I think all the people on here are super psyched about the stuff that you're making.

(06:03:03): Um, can you just give us a little bit of a intro on you?

(06:03:05): Like maybe people for people who haven't like met you online before or heard of,

(06:03:10): heard of what you're working on.

(06:03:10): Like, tell us about, tell us about yourself.

(06:03:13): And all the tool calls you introduced as well, because people need to know.

(06:03:18): Yeah, so I'm Tarek.

(06:03:20): I used to be a founder, ran a YC company for about five years.

(06:03:25): It was in video games.

(06:03:27): And then about a year ago, I was kind of wound down that company.

(06:03:31): I was trying to figure out what to do next.

(06:03:32): And yeah, it was just deeply in AI.

(06:03:36): I think they call it having AI psychosis right now.

(06:03:38): That's kind of a new term, but definitely...

(06:03:42): Yeah,

(06:03:43): and so I was,

(06:03:44): yeah,

(06:03:45): and,

(06:03:45): you know,

(06:03:45): Cloud Code just seemed like,

(06:03:46): as soon as Opus 4 came out in particular,

(06:03:49): I was like,

(06:03:49): oh,

(06:03:49): this is like,

(06:03:50): you know,

(06:03:50): I need to work on this.

(06:03:52): And so, in particular, what I do on Cloud Code is, like, I do a lot of, essentially, like,

(06:04:01): content, talking to customers, and engineering is a part of that, I think.

(06:04:06): Basically because I think for a dev tool,

(06:04:08): you really need to deeply,

(06:04:11): and especially for Cloud Code,

(06:04:12): it changes so fast.

(06:04:13): You really need to deeply know how it works at every level.

(06:04:18): And then also, I think something that we care a lot about on Cloud Code is

(06:04:23): like the feedback loop between customers and users right and just like uh what is

(06:04:28): like you know what are people saying like how can we sort of amalgamate the like

(06:04:33): you know the the feedback that we get into into things we do um how do we

(06:04:36): communicate when things are you know cloud code feels dumber or something like that

(06:04:40): so um i do all of that work and then yeah i think like um

(06:04:45): probably the biggest impact like engineering work I've done is like the ask user

(06:04:50): question tool,

(06:04:50): which was like,

(06:04:52): you know,

(06:04:52): like the interface that Claude pulls up when it wants you to ask a question and it

(06:04:59): happens,

(06:05:00): especially in plan mode,

(06:05:01): but you can kind of use it whenever I've been using it a lot to like write specs

(06:05:04): and things like that.

(06:05:05): And I think it's just one of those examples of like the models have a lot of

(06:05:09): inherent capabilities.

(06:05:11): They have a lot of like,

(06:05:13): you know, knowledge that like maybe we're not pulling out properly, right?

(06:05:18): And an anthropic,

(06:05:19): we call it like unhobbling the model very often,

(06:05:21): where it's like you have like,

(06:05:24): you're asking it to do something,

(06:05:25): but you could ask it to do something else or give it like more space in the box,

(06:05:30): kind of like put more things that it can do into it.

(06:05:33): Um, and I think that is like a forever, you know, like problem.

(06:05:37): That's kind of the goal of the cloud code team,

(06:05:39): you know,

(06:05:39): so like on hobble cloud,

(06:05:40): as much as we can,

(06:05:41): um,

(06:05:42): matching its capabilities so that it can like,

(06:05:45): you know,

(06:05:45): do the work it needs for you.

(06:05:46): So.

(06:05:47): I love that.

(06:05:48): I'm super curious.

(06:05:49): There's so many things we could talk about.

(06:05:51): Um, but, uh, and I just clicked the wrong comment to display.

(06:05:54): Okay, bro.

(06:05:55): Sure.

(06:05:55): Yeah.

(06:05:56): There's so many things we could talk about.

(06:05:58): Well,

(06:05:59): one of the things I'm really interested in is this idea that you're responsible for

(06:06:04): gathering all of the feedback and then figuring out like,

(06:06:06): how does that translate into the product that you build?

(06:06:08): How are you guys doing that in like a cloud native, AI native way?

(06:06:13): yeah so i mean i'm not the only person responsible for that i think like uh one of

(06:06:16): the things we sort of like you know at cloud code there's no one there's no qa team

(06:06:22): there's not there's a support team but like really the engineers are the support

(06:06:26): team like the engineers read the getup tickets you know things like that and so i

(06:06:30): think like this is part of sort of the evolution of

(06:06:34): you know, like agentic coding and or like how you think about engineering as your job.

(06:06:37): Like,

(06:06:38): I think it used to be that there would be like dedicated people who are like,

(06:06:41): hey,

(06:06:41): this is the feedback from users like coming down from the mountain and being like,

(06:06:45): hey,

(06:06:46): now implement it,

(06:06:47): right?

(06:06:47): And I think instead it's like,

(06:06:50): I think becoming so that everyone's roles are sort of like morphing kind of so that

(06:06:55): every engineer is in some way responsible for understanding.

(06:06:58): If you ship something you need to know,

(06:07:00): like what people think of it.

(06:07:02): You have to like address the GitHub issues.

(06:07:05): We have a lot of internal feedback as well and things like that.

(06:07:07): So I think like I think that's a really important part of what we do.

(06:07:12): And just like the culture reset is just like no one is just sort of like coding all day.

(06:07:17): And of course, Claude helps with that a lot because

(06:07:21): You know, you can like cloud is doing so much of the coding now.

(06:07:24): So you have more ability to context switch and things like that.

(06:07:28): But in a cloud native way, I think that there is like, there's something I go back and forth on.

(06:07:33): We have a bunch of like issues on like,

(06:07:35): like bots on GitHub that we like sort of,

(06:07:37): you know,

(06:07:37): scrape and things like that.

(06:07:40): And I think that like we do that, we,

(06:07:43): don't have this for Twitter exactly.

(06:07:45): And maybe I will build it.

(06:07:47): I think there's like,

(06:07:48): it's like kind of a to-do list item,

(06:07:50): but I think there's also this like level of like feel kind of just like,

(06:07:55): you know,

(06:07:55): like what,

(06:07:56): what is the feel of what people are saying?

(06:07:58): And if you get too abstracted from it,

(06:08:00): like even if Claude is doing too much work from you,

(06:08:02): I think like that can also be like hard to like understand what's like the,

(06:08:08): the meta pattern.

(06:08:09): Right.

(06:08:09): And, and so, um,

(06:08:12): Yeah, I think like when you see an issue, it's like, is this like a standalone issue?

(06:08:16): Is this like, does this pattern match against other things that you've seen?

(06:08:20): And is this like a higher root cause?

(06:08:22): You know what I mean?

(06:08:23): It's like some of the tricky stuff with feedback.

(06:08:26): But I do think that like, specifically on like,

(06:08:30): channels where there's less structured information like github github is really

(06:08:33): structured uh other channels are less structured we could like there's more to do

(06:08:37): there so basically what i hear you saying is there is some part of it but then

(06:08:42): there's this other part of it that you called feel but maybe we could call vibes

(06:08:45): that is like staying directly connected to the actual ground truth of what people

(06:08:50): are saying that's really that's really important

(06:08:53): um and and then synthesizing that into some like meta insight for yourself about

(06:08:57): okay what are the problems where are people going what's the root cause like all

(06:09:00): that kind of stuff is that a fair summary yeah i think so i think there's like i

(06:09:04): mean this is a founder lesson you learn as well like there's no replacement for

(06:09:08): talking directly to a user you know i mean like you could hire a pm and the pm's

(06:09:12): job is talk to a user and like claude could do that job for you too like it could

(06:09:16): be like oh like claude's job is to talk to the user now but like there's definitely

(06:09:20): like

(06:09:21): Um,

(06:09:21): something about like hearing directly that motivates you to do it and helps you

(06:09:24): like figure out the,

(06:09:25): like the common pattern.

(06:09:26): Like what is the,

(06:09:27): especially for something like cloud code where we think of a lot in composable

(06:09:31): building blocks,

(06:09:32): you know,

(06:09:32): like what is the composing building block that can unify a few of these different

(06:09:37): requests.

(06:09:38): Right.

(06:09:39): And,

(06:09:39): um,

(06:09:40): yeah,

(06:09:40): for that reason,

(06:09:41): like,

(06:09:42): uh,

(06:09:42): you have to sort of like be very close to the feel like,

(06:09:45): or the vibe,

(06:09:45): like you said.

(06:09:46): Um, but yeah, I think.

(06:09:49): Claude is getting better and better, and so excited to see how we can use it more and more.

(06:09:53): I want to pass this to Kieran in a sec, because I'm sure he has a question to you.

(06:09:57): But one question on my mind is...

(06:10:00): Why do you think Claude Code is having a moment right now?

(06:10:03): And why do you think it captured the zeitgeist in a way?

(06:10:05): It's not the first AI coding tool, right?

(06:10:08): And I think I have my own opinion.

(06:10:09): I'm sure Kieran does too.

(06:10:10): But I'm sort of curious for you,

(06:10:12): why do you think it's sort of like hit and is owning this moment?

(06:10:17): And what does that feel like to be riding that wave?

(06:10:21): Yeah,

(06:10:21): I mean,

(06:10:21): I think like it's a little bit like asking of like fish what water feels like a

(06:10:27): little bit,

(06:10:28): you know what I mean?

(06:10:28): Because we're like, my life is just always cloud code all the time.

(06:10:31): And so like, you know, some people are like, oh, like there's more cloud code now.

(06:10:34): I'm like, oh, I haven't noticed.

(06:10:36): You're like, welcome to my world.

(06:10:37): Yeah, yeah, yeah.

(06:10:39): So I think there's a little bit of that.

(06:10:41): I think.

(06:10:42): uh it's always hard to attribute this this stuff like that's kind of the difficulty

(06:10:46): of like the vibe right i would say like what is truly unique about cloud code is

(06:10:51): that everyone talks about startup within a startup you know like oh like this is

(06:10:55): like you know the classic pitch that felt like a big company gives founders to come

(06:10:59): like work for them but cloud code is like truly that you know i mean like and i

(06:11:04): think that means like we clearly clearly care about this product in a way that like

(06:11:09): Like, that's the moat, you know what I mean?

(06:11:10): Like,

(06:11:11): like,

(06:11:11): like Boris is like an incredible,

(06:11:14): true founder within Anthropic,

(06:11:15): kind of like,

(06:11:16): you know,

(06:11:16): I mean,

(06:11:16): like a founder of cloud code within Anthropic.

(06:11:18): And he,

(06:11:18): and,

(06:11:19): you know,

(06:11:19): he really gets the ability to run with that and like,

(06:11:22): make it like an incredible product.

(06:11:24): And he recruits other people who are like,

(06:11:27): yeah,

(06:11:28): ex-founders,

(06:11:28): high agency,

(06:11:29): just like we have an incredible engineering team.

(06:11:33): of the incredible product team.

(06:11:35): And we really are very bottoms up.

(06:11:38): And I think that everyone just cares a lot.

(06:11:44): And just in every part of the product, you feel that.

(06:11:50): And I think also in how

(06:11:52): we think about like improving it you know i mean like i think we try and always be

(06:11:56): the first one to like come up with something new you know i mean and if like people

(06:12:00): are like following up on us that's great that's like you know i think in anthropic

(06:12:04): we call it the race to the top right like we want to make sure that we're like sort

(06:12:06): of setting the standard of what is like an amazing product an amazing developer

(06:12:10): experience right um and so yeah that's just like i

(06:12:15): Like,

(06:12:16): you know,

(06:12:16): to me,

(06:12:16): it's sort of like this the exceptional thing is just like that this exists,

(06:12:22): you know,

(06:12:22): and that we're like grinding at it.

(06:12:24): And then, of course, Opus is also like incredible.

(06:12:27): Right.

(06:12:27): And like these like two things like we have such an excellent research org and we

(06:12:32): have such an excellent product org that

(06:12:35): That's the combination I think would make it inevitable.

(06:12:38): Is there some sort of collaboration between the two that you think makes it like,

(06:12:43): is it actually that important for them to be in the same company?

(06:12:45): And is there like a good overlap there?

(06:12:48): Like, how do you guys do that?

(06:12:49): Yeah,

(06:12:50): I mean,

(06:12:50): like,

(06:12:52): we generally don't talk too much about,

(06:12:54): like,

(06:12:54): you know,

(06:12:54): how we,

(06:12:57): like,

(06:12:57): you know,

(06:12:58): like...

(06:12:58): It's just us friends,

(06:12:59): though,

(06:13:00): you know?

(06:13:00): Yeah, exactly.

(06:13:01): Us and, like, a thousand of our friends.

(06:13:03): 16,000, yeah.

(06:13:03): Yeah, yeah, exactly, exactly.

(06:13:05): Yeah, I think that, like...

(06:13:10): like you know obviously anyone can look at the cloud code like sdk or like you know

(06:13:15): you can like reverse the build and you're like oh this is just calling the api and

(06:13:19): like i could call the api so why don't i call the api better than you call it you

(06:13:22): know i mean like um

(06:13:26): Yeah, maybe in an alternate world, Cloud Code is built somewhere else.

(06:13:29): I do think one thing we do at Anthropic that you see very consistently is just we

(06:13:34): hire people who really believe that the models are going to get better.

(06:13:37): And I think what that actually means is a little bit scary, you know what I mean?

(06:13:44): like they've already gotten so good and they'll keep getting better and so we look

(06:13:47): at a feature sometimes like skills and we're like oh this is like kind of good now

(06:13:51): but what if it was great you know i mean and like i think we're always like that

(06:13:55): question of like unhobbling the model you to unhobble the model you have to believe

(06:13:59): that the model will get better and better and like um if you don't believe that

(06:14:03): then you're like oh like let me think of like this more complicated orchestrated

(06:14:07): solution or something you know and uh i think anthropic just like hires those

(06:14:11): people and

(06:14:12): gives us the ability to do it.

(06:14:15): Amazing.

(06:14:15): Kieran, what's on your mind?

(06:14:17): Yeah,

(06:14:18): so basically you're saying my plugin,

(06:14:21): the Compound Engineering plugin,

(06:14:23): is just too much,

(06:14:25): and the model will get better,

(06:14:26): and then we don't need it.

(06:14:27): No, I'm very curious.

(06:14:29): So I have a Compound Engineering plugin that I use on top of Cloud,

(06:14:34): which is how I build software,

(06:14:36): because

(06:14:38): whenever I started using closed codes on 3.7, it was wild.

(06:14:44): And it went anywhere.

(06:14:45): And it needed something.

(06:14:48): And over the time, you've been

(06:14:53): like re-releasing versions, better models of that.

(06:14:56): And I've been deleting parts of this.

(06:14:58): Like I don't need always to go in my planning mode.

(06:15:01): You can just also use your planning mode for smaller things where I don't need.

(06:15:06): And that's great.

(06:15:07): So I'm curious,

(06:15:08): like from all the moving parts,

(06:15:11): like I have planning mode,

(06:15:12): I have work,

(06:15:12): I have like testing,

(06:15:14): like some brainstorming.

(06:15:18): I'm sure you're thinking about bringing things

(06:15:21): these things into Cloud Code as well.

(06:15:25): And what are these building blocks?

(06:15:29): For example, Skills, I love Skills.

(06:15:31): I now get it.

(06:15:32): Whenever Skills launched first,

(06:15:33): I was like,

(06:15:33): yeah,

(06:15:34): Just-In-Time is cool and all that,

(06:15:36): but what does it mean?

(06:15:38): And now it really clicks because that's the composability layer that we kind of

(06:15:44): need,

(06:15:45): especially in bringing your own version of something.

(06:15:50): So what are these things that you're,

(06:15:52): like,

(06:15:52): hearing from people that you're trying to capture in,

(06:15:55): like,

(06:15:55): a nice form?

(06:15:57): Like, can you share some of these things?

(06:16:00): Yeah, I mean, I think there is maybe, like, a meta question here that I can sort of talk about.

(06:16:06): Like,

(06:16:07): you know,

(06:16:07): you're talking about,

(06:16:07): like,

(06:16:07): combat engineering and the,

(06:16:08): like,

(06:16:09): concepts you're building and,

(06:16:10): like,

(06:16:11): yeah,

(06:16:11): like,

(06:16:12): how do we think about,

(06:16:12): like,

(06:16:13): sort of adding these concepts or,

(06:16:15): like,

(06:16:15): you know,

(06:16:15): will the models get better?

(06:16:16): And I think that, like...

(06:16:18): It's a very delicate balance.

(06:16:20): You know, this is like oftentimes like it's an art, not a science to me.

(06:16:24): You know what I mean?

(06:16:24): Of like, what does building an agent loop feel like?

(06:16:27): You know, it's like you need to really just be very like tightly intertwined with it.

(06:16:33): There's also the UX questions of like, okay, like it slash command plan mode.

(06:16:36): Like, you know, like how does the user know how to like interact with the model?

(06:16:40): and how does the model help that?

(06:16:42): Like we rolled out prompt suggestions recently,

(06:16:44): it was like a first stab at trying to help the model help you,

(06:16:47): you know?

(06:16:48): But there's so much more that can be done there, you know?

(06:16:52): I do think that like, this is true of all AI products.

(06:16:55): It's like the AI capabilities are like here,

(06:17:00): but with some engineering work,

(06:17:01): you can make it like go up to here and then the next model comes out and it's here

(06:17:05): and you have to like delete the code.

(06:17:07): You have to delete the code and then you have to restart again.

(06:17:10): And most people get stuck at deleting the code.

(06:17:12): You know what I mean?

(06:17:13): And they're like, oh, no.

(06:17:15): i don't want to you know like like i don't believe the model is this good yet or

(06:17:19): not right um and i think that like uh that's you know um where like i guess i can

(06:17:27): preview it on stream we're like uh we're replacing to do's with tasks right now

(06:17:32): right we're like to do's are like a you know sort of like an outdated system and

(06:17:36): we're like you know upgrading it to be work better for long context and multi-agent

(06:17:40): flows um and that's like an example of us like

(06:17:44): trying to keep with the model and unhovel it.

(06:17:49): But yeah, I think the stuff you do right now is still great.

(06:17:55): Anything that you can find to make yourself more productive right now is amazing.

(06:17:58): You just have to reevaluate.

(06:17:59): Is this still serving me as the models get better?

(06:18:02): Or do I delete the code?

(06:18:03): Do I delete the thing?

(06:18:04): Can I turn this into something simpler?

(06:18:05): Can I turn it into a skill?

(06:18:07): Can I turn it into something that the model can use a little bit more natively?

(06:18:13): um like for me for example like uh something that i really started liking for opus

(06:18:17): 4.5 was asking opus to like interview me um and like that i had tried with that

(06:18:23): with sonnet 4.5 you know and like it just wasn't as good and then opus 4.5 was like

(06:18:28): oh like it was incredible um and like i think you just have to like really that's

(06:18:34): the hard part about ai engineering is like you have to like retry these things

(06:18:38): delete code like see what's working what's not and um

(06:18:43): Yeah, like, you just have to, like, it's kind of an art there, I think.

(06:18:47): There's a couple things in there that I thought were amazing that I really want to dig into.

(06:18:51): The first one is the obvious one.

(06:18:53): Like, you said you're thinking about replacing to-dos with tasks.

(06:18:56): Can you explain, like, what a task would be and how that differs?

(06:19:00): Yeah, so, you know, I literally have, like, an article that we'll be going up in a bit.

(06:19:06): But a task... You heard it here first, folks.

(06:19:09): Tasks coming to... Exclusive!

(06:19:10): Exclusive!

(06:19:13): um yeah please don't tweet this yet like i will uh you know um i'll post it uh but

(06:19:18): yeah and you were inspired by beads and uh steve's work there um and i think we're

(06:19:23): just like uh you know adding dependencies between graphs and also making it so that

(06:19:28): these can persist across multiple sessions across multiple agents you know so so a

(06:19:33): task is like rather than being something that's ephemeral that only one claude

(06:19:37): knows about it's something that

(06:19:39): is meant to be passed between different clauses and can persist between different sessions.

(06:19:44): Yeah, and then also it's more project-based.

(06:19:47): A to-do list tends to be kind of flat,

(06:19:50): and we just basically see Claude sort of limit itself when it just has a flat list.

(06:19:54): Yeah, like in AMP, they have threads, I think, something like that.

(06:19:59): And the current task is now just a sub-agent.

(06:20:05): Will that be replaced as well by this?

(06:20:08): Oh, yeah, I see.

(06:20:09): Yeah, this is kind of... I believe...

(06:20:12): We've renamed like... Sup agents?

(06:20:16): We've renamed task, I think.

(06:20:18): I don't think that's the actual name.

(06:20:19): I think it's agent.

(06:20:21): Okay, maybe that's now a skill.

(06:20:22): I don't know.

(06:20:22): Yeah.

(06:20:23): Yeah, yeah, yeah.

(06:20:24): I need to double check on this.

(06:20:26): But yeah, like tasks are definitely like to-dos like V2.

(06:20:29): Yeah.

(06:20:30): Yeah.

(06:20:31): And so as you're building this,

(06:20:33): I know you guys do a lot of dogfooding,

(06:20:36): or I think you call it antfooding internally.

(06:20:38): Can you tell us about that as this sort of product development loop for yourself

(06:20:43): and how that has worked in this case,

(06:20:45): for example?

(06:20:46): yeah i mean i think it's like the classic sort of like it's not too different from

(06:20:51): like i think what any other company can do in the sense of like you know you ship

(06:20:55): product to your company that they use and they give you feedback um and i think one

(06:21:01): thing special about cloud code is obviously we use it ourselves and so like you

(06:21:04): know at any one point i usually have like some sort of like vibe coded experiment

(06:21:08): that i'm

(06:21:09): running on my cloud code to do other things and see if,

(06:21:12): like,

(06:21:12): oh,

(06:21:12): does this make it better or not?

(06:21:14): And, like, 90% of the time, it doesn't make it better.

(06:21:16): And it's kind of, like, worse, right?

(06:21:19): But so the first filter is, like, your own judgment.

(06:21:22): And then, yeah, like, obviously, like, you know, like, it just comes down to caring again.

(06:21:28): You know, like, we really, really care about every piece of feedback.

(06:21:31): We really, like, if someone on Anthropic is, like, hey, this is great, like, you know, we, like,

(06:21:37): we'll just respond and you know i mean i think like that's uh it's not anything

(06:21:41): special there you know it's just like caring about the feedback the special thing

(06:21:45): that i i heard you say that is not necessarily ant fooding but like the another

(06:21:49): special thing that you just said that i really it's like it's something that i'm

(06:21:52): trying to figure out how do we do this inside of every and so you sort of like blew

(06:21:56): my mind because it's like you gave me a way to think about it is yeah okay one is

(06:22:00): the idea of a race to the top

(06:22:02): Um,

(06:22:03): we're trying to make the best quality product,

(06:22:05): uh,

(06:22:05): we possibly can with the most powerful AI that we possibly can.

(06:22:08): How do we do that?

(06:22:08): And it seems like there's this cycle,

(06:22:11): uh,

(06:22:12): that every AI company runs into,

(06:22:14): which is like,

(06:22:15): okay,

(06:22:15): you're building something.

(06:22:16): You want the model to do it.

(06:22:17): So you write a bunch of code to like help the model do it.

(06:22:20): And then in three months, the model is better.

(06:22:22): And you're like, fuck, like, what do I do with all this code that I just wrote?

(06:22:25): And I think a lot of people,

(06:22:27): like you,

(06:22:27): like you said,

(06:22:28): a lot of people,

(06:22:29): you kind of almost want to ignore it because you're like,

(06:22:32): i don't want to throw out this product but um we're in this like weird place where

(06:22:37): we're all racing to the top and at some point probably like

(06:22:42): Uh, this will, this, I would, I imagine this will settle down a bit.

(06:22:45): It won't be like every three months we will have to like redo all of our products,

(06:22:48): but at least for now,

(06:22:49): it's like a,

(06:22:50): um,

(06:22:51): we accept that every three months where it's,

(06:22:53): it's a practice that we go through our product and like delete a bunch of stuff and

(06:22:57): like re architect it.

(06:22:58): And, and that's actually okay.

(06:23:00): And it's not just delete a bunch of stuff and you have the same functionality.

(06:23:03): It's you delete a bunch of stuff.

(06:23:04): The model takes care of a lot of the things you used to use code for,

(06:23:07): and then you use code to figure out,

(06:23:08): okay,

(06:23:08): what's the next thing that we can do that that's the sort of next.

(06:23:11): reach to the top is that kind of the the framework you guys have um yeah i think

(06:23:17): that like roughly like you know if cloud is writing 10 times as much code you can

(06:23:22): also delete 10 times as much code you know like there's also that i didn't even

(06:23:25): include that in my analysis yeah yeah so i think like the the big thing here is

(06:23:29): like javon's paradox kind of right like you can write more and more software

(06:23:32): everyone can write more and more software right and so uh you need to like uh like

(06:23:37): the

(06:23:38): bar of quality is very very high right like you have to really deliver like just an

(06:23:42): excellent excellent product and um yeah like if not like someone will vibe code

(06:23:46): like a clone of your product like right away you know what i mean uh like and and

(06:23:50): so that's the other thing is like i think when you think about you know like

(06:23:55): moats and like how you're building things like obviously anyone can look at a

(06:23:59): single point you have and be like oh like at that point i'm going to copy that

(06:24:02): point but they can't copy the vector right like they can't copy like the directions

(06:24:05): that you're going in right um and then there's also a lot of learnings and what was

(06:24:09): thrown out you know i mean like i think with the ask user question tool for example

(06:24:12): i've seen a lot of different clones across different like um like coding agents and

(06:24:17): it's curious to me like what they don't

(06:24:20): what I threw out that they didn't.

(06:24:22): You know what I mean?

(06:24:24): Also, you're the creator of that tool.

(06:24:26): You're the inventor, creator.

(06:24:28): Thank you so much.

(06:24:29): It's a great tool.

(06:24:30): Yeah, yeah.

(06:24:31): I mean, I was surprised at how good it was or how much people liked it as well.

(06:24:37): Yeah, that's just also, like, the most concrete example I have of this.

(06:24:41): And, like, I think, yeah, like, that's, like, the decisions you make.

(06:24:46): Like, now I have all this context about, like, what is it like to create better elicitation?

(06:24:51): Like, what do I want the next models to be able to do to make this tool better?

(06:24:55): You know, like, or when do I delete this tool?

(06:24:56): You know what I mean?

(06:24:57): Like, is, like, and I have all this context.

(06:24:59): And, like, that's, like, a really hard thing to clone or copy, right?

(06:25:03): And, again, like, a moat of, like,

(06:25:07): how you think about building software companies these days.

(06:25:10): I feel like Cloud Code created this movement back towards the terminals and away from like IDEs.

(06:25:17): But on this stream,

(06:25:18): a few people,

(06:25:19): including me,

(06:25:20): I've had a little bit of this sense,

(06:25:21): but there's a couple of people who are more strong,

(06:25:23): more strongly have this sense that we've kind of speed run the terminal era and

(06:25:29): we're going back to GUIs now.

(06:25:30): Like people, for example, are using Conductor would be a good public example.

(06:25:34): What do you think?

(06:25:34): Does the terminal have legs?

(06:25:36): Is it going to last?

(06:25:37): Or are we moving back to GUIs?

(06:25:40): Yeah, what's your thought?

(06:25:44): Yeah, no idea.

(06:25:44): I think, like, we will, you know, like, let's race to the top and find out.

(06:25:49): You know,

(06:25:49): like,

(06:25:50): I think,

(06:25:50): like,

(06:25:51): we are obviously doing,

(06:25:52): you know,

(06:25:53): a bunch of different UI things with Cloud Code Remote and Cloud Code or Cloud Code

(06:25:56): Web and Cloud Code Desktop and obviously Cowork and,

(06:26:01): like,

(06:26:01): all of these things.

(06:26:02): And so there's lots of different UIs.

(06:26:04): I think there is something about, like, the...

(06:26:06): keyboard first interface of terminal and like just like the simplicity and like uh

(06:26:11): the responsiveness of it that is like a high watermark to beat you know um but yeah

(06:26:19): i don't like um i think we just have to like yeah care a lot iterate a lot and see

(06:26:24): what see what happens yeah sweet karen any questions

(06:26:29): Do you run a Mac Mini?

(06:26:31): Because we heard a lot of people say that they run a Mac Mini.

(06:26:34): Oh, for dangerously skipped permissions?

(06:26:36): No, just like full-time running Claude doing stuff for them.

(06:26:41): Yeah, you know, okay, um...

(06:26:45): I feel like maybe people will be disappointed at how much,

(06:26:48): like,

(06:26:48): you know,

(06:26:50): like,

(06:26:50): we're not doing that,

(06:26:52): I think.

(06:26:52): Some people end up off the car, I think.

(06:26:54): That's fine.

(06:26:55): Like, I'm just curious, like, if you guys do that.

(06:26:58): You said you were at the edge.

(06:27:00): You said you were going to the top.

(06:27:01): Like, you gotta have a back game.

(06:27:05): I think that it's like long running task.

(06:27:07): I think it's still, to me, the question is like, is it producing economically viable output?

(06:27:14): You know what I mean?

(06:27:14): Like, are you creating something that is truly useful?

(06:27:18): How much in the loop do you need to be, you know?

(06:27:20): And like, how much ambiguity do you need to reduce?

(06:27:24): And so like,

(06:27:25): for example,

(06:27:25): like the spec mode stuff,

(06:27:27): I actually felt helped me reduce the ambiguity of a feature quite a lot so that

(06:27:32): then I could have Opus run for long periods of time.

(06:27:34): but it took a lot more work out of me like you know like i would now like i'm

(06:27:38): writing like specs for like 30 minutes you know and then like then that cloud code

(06:27:42): is working right and but it's like um i'm still like putting in a lot of work i'm

(06:27:47): not just sort of like telling cloud to go do something so

(06:27:50): And yet,

(06:27:51): Opus 4 or 5 is very,

(06:27:53): very good at doing these longer thinking,

(06:27:57): bigger things,

(06:27:58): not making mistakes and self-correcting,

(06:28:00): yeah.

(06:28:01): That's right.

(06:28:02): Yeah, that line will keep going.

(06:28:07): Are you also struggling with Cloud Code addiction like the rest of us?

(06:28:11): um what's your recommended uh strategy for that he's like at home i don't do

(06:28:16): anything it stays at work no i mean that's what i was supposed to talk about right

(06:28:21): like was like stuff i do at home when i was cloud code um i think uh i i i use

(06:28:28): cloud code a lot for sure yeah yeah um i think um

(06:28:33): Yeah,

(06:28:34): I definitely,

(06:28:36): yeah,

(06:28:36): like there's this mix of like,

(06:28:38): what am I doing for work that's sort of like fun versus like,

(06:28:42): like,

(06:28:43): you know,

(06:28:43): for fun,

(06:28:44): fun.

(06:28:44): But yeah, I'm struggling just like you.

(06:28:47): We can start doing an anonymous group or something.

(06:28:50): Sweet.

(06:28:51): Dark, it was really great to have you on.

(06:28:54): Thank you so much for joining and for taking the time.

(06:28:55): If people want to reach you, where can they find you?

(06:28:59): Yeah, I'm on Twitter, at TRQ212.

(06:29:02): Yeah, post a lot there.

(06:29:05): Yeah, definitely tag me with, like, feedback and stuff.

(06:29:07): And,

(06:29:09): yeah,

(06:29:09): just wanted to say thank you to you guys,

(06:29:10): too,

(06:29:10): for,

(06:29:11): like,

(06:29:11): you know,

(06:29:11): I think AI coding and AI overall is,

(06:29:15): like,

(06:29:15): actually a really kind of scary subject to a lot of people.

(06:29:18): And I think a lot of people have a lot of anxiety around it,

(06:29:20): you know,

(06:29:20): and understandably,

(06:29:22): you know,

(06:29:22): like,

(06:29:22): I think Dario said a lot about,

(06:29:24): like,

(06:29:24): how...

(06:29:25): we would like this to go a little bit slower if we can, you know?

(06:29:29): But, you know, we're, we are where we are and we just want to race to the top.

(06:29:33): We want to like show people, everyone, everything that like can be done.

(06:29:37): We want to like bring people along for the ride.

(06:29:40): And I think you guys are doing like really important work for like,

(06:29:42): you know,

(06:29:43): making AI more accessible to people.

(06:29:44): So.

(06:29:45): Thank you.

(06:29:46): I really, really appreciate that.

(06:29:47): Thank you for joining.

(06:29:48): Well, let's, let's talk soon.

(06:29:50): That's good.

(06:29:50): Thank you.

(06:29:52): All right, folks, you heard it here from Tharik.

(06:29:55): If you want to stay at the edge of AI, every is the place to do it.

(06:30:03): I actually really truly actually means a lot.

(06:30:05): We spend a lot of time thinking about how anxious people tend to feel and how fast

(06:30:13): things are moving and how to create a place inside of every where you have

(06:30:17): everything that you need to not only stay on top of things,

(06:30:21): but actually build a really great life and really great business.

(06:30:24): Everything from all the ideas that we publish to all the apps that we build to the

(06:30:27): trainings that we do.

(06:30:28): It's with that in mind so that you can understand what's happening and then use it

(06:30:34): to build a better life.

(06:30:36): With that, I'm welcoming back our COO, Brandon.

(06:30:40): Brandon, how are you doing?

(06:30:40): Yeah.

(06:30:41): i'm good man how are you i'm doing well um and naveen who is the uh gm gm of