WEBVTT NOTE Sentence-level transcript of https://www.youtube.com/watch?v=qqrk7CtkuIw NOTE One cue per sentence. Cue ids are the line anchors on /transcripts/qqrk7CtkuIw.html. A cue ends where the next begins, or 2 s after its last word. s1 00:00:01.309 --> 00:00:03.309 [music] s2 00:00:12.720 --> 00:00:15.720 Joining us on stage is the co-founder of Instagram s3 00:00:15.720 --> 00:00:16.280 [music] s4 00:00:16.280 --> 00:00:19.760 and a member of technical staff at Anthropic. s5 00:00:19.760 --> 00:00:22.120 Mike Krieger. s6 00:00:36.360 --> 00:00:37.360 How's everybody doing? s7 00:00:37.360 --> 00:00:39.160 I mean, good morning. s8 00:00:39.160 --> 00:00:40.520 Nice. s9 00:00:40.520 --> 00:00:41.640 Um s10 00:00:41.640 --> 00:00:44.520 Mike, thank you for releasing Fable just in time for us. s11 00:00:44.520 --> 00:00:45.640 Exactly for the conference. s12 00:00:45.640 --> 00:00:46.505 We timed it. s13 00:00:46.505 --> 00:00:47.280 [laughter] s14 00:00:47.280 --> 00:00:49.760 Um we're we're so glad to have you. s15 00:00:49.760 --> 00:00:55.600 Uh you're uh one of the preeminent builders and you're a leading labs at Anthropic. s16 00:00:55.600 --> 00:01:02.280 Um how has your model usage changed as as you've, you know, seen models internally grow? s17 00:01:02.280 --> 00:01:05.800 Yeah, I mean, for me it's been like both the model shift and then my role shift. s18 00:01:05.800 --> 00:01:09.120 So, I for like the first 2 years I was at Anthropic, I was chief product officer. s19 00:01:09.120 --> 00:01:16.720 And then I kept seeing people build with the models and the FOMO just kept increasing because I was you know, use the models as much as possible. s20 00:01:16.720 --> 00:01:26.480 But for example, on product strategy I would write a strategy doc and then have Claude critique it and maybe you can use a workflow, but it's not quite the same as like building in that pure way. s21 00:01:26.480 --> 00:01:30.920 And I was like spending all my weekends trying to build with it and I realized, "Okay, I actually just need to shift. s22 00:01:30.920 --> 00:01:32.200 It's like way too interesting a time." s23 00:01:32.200 --> 00:01:41.000 And it's actually an interesting trend I've seen now like several people that were CTOs at other places are like now joining as ICs at Anthropic and other places. s24 00:01:41.000 --> 00:01:49.400 But I made a role shift and it was actually right around the time where we started getting sort of internal snapshots of what became Mythos and Fable. s25 00:01:49.400 --> 00:02:02.200 And what was really interesting watching that sort of shift was um that kind of change between I have an idea, I'm going to like sort of break it down in my head much more how I would do engineering normally, s26 00:02:02.200 --> 00:02:13.720 and then kind of iterate through these different steps to moving to much more of the paradigm of I'm going to describe the goal, like go off and work on it, and then like we can talk about what trade-offs you you know, surface some questions along the way, but then s27 00:02:13.720 --> 00:02:16.520 figure out what where you landed and where we can go from there. s28 00:02:16.520 --> 00:02:17.160 I find it's hard. s29 00:02:17.160 --> 00:02:21.800 I don't know if people have this experience where people people's only been re-enabled for a couple of days. s30 00:02:21.800 --> 00:02:23.680 People's definitely way way smarter than me. s31 00:02:23.680 --> 00:02:26.280 So, sometimes it'll finish work and be like, here's the trade-offs I made. s32 00:02:26.280 --> 00:02:32.440 I'm like, can you explain it to me like I'm a little dumber than you are because I need you to like sort of break this down for me. s33 00:02:32.440 --> 00:02:40.240 But, that's been one sort of big change is sort of moving from that task delegation to like express the end state and then have it go and and cook on it. s34 00:02:40.240 --> 00:02:43.560 Yeah, we're all learning how to delegate better. s35 00:02:43.560 --> 00:02:46.360 Tariq did us a huge favor yesterday. s36 00:02:46.360 --> 00:02:50.240 We Did you want to read it in the newspaper? s37 00:02:50.240 --> 00:02:54.840 You know, that we have we have write-ups of talks now in in like the next day's newspaper. s38 00:02:54.840 --> 00:02:57.080 He said, be unreasonable. s39 00:02:57.080 --> 00:02:59.880 In what ways you know, you know, have you been more ambitious? s40 00:02:59.880 --> 00:03:00.920 Yeah, you're prompting. s41 00:03:00.920 --> 00:03:03.200 I love that I I mean, I love that framing. s42 00:03:03.200 --> 00:03:04.840 We actually just hit this today. s43 00:03:04.840 --> 00:03:14.320 I'm one of the labs initiatives I have is internal product, and somebody was like, hey, it doesn't work the way I want it to, and can you make some changes? s44 00:03:14.320 --> 00:03:17.480 And I realized, oh, I'm just going to go ask Claude to do this. s45 00:03:17.480 --> 00:03:18.720 Like, why don't you ask Claude? s46 00:03:18.720 --> 00:03:20.480 And this was a non-technical person. s47 00:03:20.480 --> 00:03:28.680 So, I actually think as an industry or even as a product team, we have to teach people to be more unreasonable in their usage, and it's sort of hard to imagine. s48 00:03:28.680 --> 00:03:37.959 I think that that if I can digress for a second on product design, I think right now the like kind of first generation of AI products, we put them too much in a box and constrain their s49 00:03:37.959 --> 00:03:43.360 their sort of access to tools or kind of degrees of freedom, which means it was much harder to be unreasonable, right? s50 00:03:43.360 --> 00:03:47.360 When you say, do this thing for me, and then it would be like, well, I can't. s51 00:03:47.360 --> 00:03:53.680 I can barely like I can write code, but I can't really run it, or I can kind of introspect my environment, but not really. s52 00:03:53.680 --> 00:04:02.920 Um and I think as you see our own like product progression even with things like co-work where like, you know, does every single like knowledge worker need a virtual machine that can write bash? s53 00:04:02.920 --> 00:04:10.480 Like, on the face of it, no, but then when you realize, oh, actually, that way it can remediate an issue where, oh, I tried to parse a PDF using our built-in PDF parser. s54 00:04:10.480 --> 00:04:13.320 I hit this yesterday and it was like, ah, I can't parse it this way. s55 00:04:13.320 --> 00:04:16.519 Well, okay, well, I can probably write a script that can do this as well. s56 00:04:16.519 --> 00:04:17.880 Um so, I think that's it. s57 00:04:17.880 --> 00:04:23.160 My most unreasonable thing though was uh one of our labs projects I wrote in Python like near and dear to my heart. s58 00:04:23.160 --> 00:04:24.720 All of Instagram was in Python. s59 00:04:24.720 --> 00:04:29.800 But I think they're finally converting it to PHP now that they have um like models that can do it. s60 00:04:29.800 --> 00:04:30.518 I know. s61 00:04:30.518 --> 00:04:31.720 [laughter] s62 00:04:31.720 --> 00:04:32.520 Tokens. s63 00:04:32.520 --> 00:04:38.120 Um and uh for deployment I realized that Cloud Code had like figured out a better deployment story with Bun. s64 00:04:38.120 --> 00:04:41.120 And I was like, okay, I need to port this whole thing from Python to TypeScript. s65 00:04:41.120 --> 00:04:46.400 Like, as a, you know, if I put on my like 2010s engineering hat or even my early 20 20s, like, that's a dumb idea. s66 00:04:46.400 --> 00:04:50.800 Like, who would ever port like, at that point, you know, a couple hundred thousands of lines of code. s67 00:04:50.800 --> 00:05:06.520 Um but I was like, I think this is doable now and I basically created this dynamic workflow setup and over the weekend had it port the whole thing, like, verify it, double-check it, then read both code like this basically churn and churn and churn and then came back Monday to a completed workflow that was a ported version s68 00:05:06.520 --> 00:05:06.960 of that thing. s69 00:05:06.960 --> 00:05:09.520 So, that probably ranks on like the more unreasonable things. s70 00:05:09.520 --> 00:05:16.280 Like, yeah, just port this entire Python code base to TypeScript, get it working, get it deployable in, you know, a weekend. s71 00:05:16.280 --> 00:05:21.000 Yeah, I mean, a lot of people are talking about the the Bun Zig to Rust version. s72 00:05:21.000 --> 00:05:27.120 I think a lot of people are also like, well, it's a compiler, it's a it's a runtime, it's got lots of tests, easy to do. s73 00:05:27.120 --> 00:05:32.640 Can you port Instagram, which you would know very well, to PHP like that? s74 00:05:32.640 --> 00:05:33.760 Like a like a product. s75 00:05:33.760 --> 00:05:37.440 Yeah, I mean, I think the product side of it it's even I don't know if it's easier or harder. s76 00:05:37.440 --> 00:05:46.560 One of the things we did at Instagram, this is when Python 3 came out and we were able to add type hints for the first time and it was people had a lot of internal conversations like, are we going to run out of steam on Python? s77 00:05:46.560 --> 00:05:54.280 And my perspective was always like, I think we can take this way further than we think we can, uh, but I think types are going to help us not sort of be in our own way. s78 00:05:54.280 --> 00:06:04.760 And we built this thing called Monkey Type where we basically like captured runtime type like basically the types that were actually getting used in production and then map those back to to the types in the code base. s79 00:06:04.760 --> 00:06:17.400 And I think because of that sort of pattern, I think there's really interesting ways in which if you're doing sort of conversion or sort of cross compiling using LLMs, you can also lean on production data a lot more or run sort of like segmented tests. s80 00:06:17.400 --> 00:06:20.040 I think that like there's a lot of, uh, things you can do there. s81 00:06:20.040 --> 00:06:22.680 But yeah, I think it's, I mean, the sky's the limit there as well. s82 00:06:22.680 --> 00:06:29.240 I think the hardest part is always finding the boundary around where you can start doing it incrementally without trying to boil the whole ocean and like swap it overnight. s83 00:06:29.240 --> 00:06:38.919 Yeah, I mean, your users are your test ultimately and, um, you know, I we I also read another article in the newspaper about how you could just use rollouts and sometimes you don't really know, s84 00:06:38.919 --> 00:06:45.560 uh, what you're going to need it for, but when that infrastructure exists for your experiments and to roll things out, it's enabled so much. s85 00:06:45.560 --> 00:06:47.360 Yeah, I mean, I always found this was advice we got. s86 00:06:47.360 --> 00:06:53.840 It's like we launched Instagram and the happened to be the first week everything melted cuz we didn't really know what we were doing on the back end side of things. s87 00:06:53.840 --> 00:06:59.480 And, uh, coincidentally that week there was like a lunch that one of our investors just scheduled like not even for us. s88 00:06:59.480 --> 00:07:01.640 It was just a infrastructure lunch. s89 00:07:01.640 --> 00:07:07.800 And we ended up spending we totally like monopolized that conversation cuz everybody had their own opinion about how we could fix our scaling. s90 00:07:07.800 --> 00:07:23.320 Um, and like the two pieces of advice I got there is like 2010 that I like will forever retain is like, um, like basically like pre-measure everything that you think you might even remotely need because the worst thing is an outage where you're like, well, is this like number normal or is it high? s91 00:07:23.320 --> 00:07:26.480 And like, oh, I don't know because I don't have data until I just added this metric. s92 00:07:26.480 --> 00:07:29.320 And the other one is being like really thoughtful about knobs and feature flags. s93 00:07:29.320 --> 00:07:36.160 So even, you know, early Instagram we had like a very, uh, simple but really effective like way in which you could do like ramp outs and rollouts. s94 00:07:36.160 --> 00:07:46.280 And dynamic config too where, you know, a lot of our runtime configurations had to be changed, you know, in a matter of seconds so that we could handle load and being able to like do that in a first class way was was really important. s95 00:07:46.280 --> 00:07:52.400 I'm seeing that definitely in in AI as well where, you know, we're making all sorts of different trade-offs and having that kind of runtime configuration is super key. s96 00:07:52.400 --> 00:07:53.240 Yeah. s97 00:07:53.240 --> 00:07:58.920 Uh my my favorite scaling story of Instagram by the way, I think it's like your launch day when you you DDoS yourself with the email. s98 00:07:58.920 --> 00:08:00.240 Yes. s99 00:08:00.240 --> 00:08:03.000 Which I people should look up that story if uh if you haven't seen it. s100 00:08:03.000 --> 00:08:04.960 Um I wanted to go into tags. s101 00:08:04.960 --> 00:08:06.800 Uh very very major shift. s102 00:08:06.800 --> 00:08:09.760 Uh it's it's how 60 something percent of your code is written today. s103 00:08:09.760 --> 00:08:10.400 Yeah. s104 00:08:10.400 --> 00:08:15.440 Um how do you square that with everything you just said where it's like very dynamic? s105 00:08:15.440 --> 00:08:18.720 Like you don't actually ship one app, you ship one app with 3,000 flags. s106 00:08:18.720 --> 00:08:19.040 Yeah. s107 00:08:19.040 --> 00:08:20.360 And like, well, what are you working on today? s108 00:08:20.360 --> 00:08:20.720 I don't know. s109 00:08:20.720 --> 00:08:23.040 Like it's it's for this segment of the population. s110 00:08:23.040 --> 00:08:23.880 Yeah. s111 00:08:23.880 --> 00:08:25.160 Yeah, I mean, I think there's a bunch of things. s112 00:08:25.160 --> 00:08:27.240 So, like with I was really excited. s113 00:08:27.240 --> 00:08:36.080 I was talking to Swix earlier like, I'm really excited that we have tag out there because it is uh how we've been working for a while and I would get up on stages and people like, "How do you work at Anthropic?" s114 00:08:36.080 --> 00:08:46.720 And I'd be like, "Oh, yeah, we use these things like that are not quite Claude code, but you know, uh but it's hard to describe it, but I mean, if you like got to poke into Anthropic, like you would see s115 00:08:46.720 --> 00:09:00.280 uh of course Claude code usage for things that are like more interactive or if you're kind of iterating on a particular uh sort of sort of specific thing where you want a lot of like a high sort of bandwidth back and forth, but most usage is actually much more delegating s116 00:09:00.280 --> 00:09:02.000 uh via tagging and via tag. s117 00:09:02.000 --> 00:09:06.040 And you can say like, "Here's the And the reason it's really interesting is how multiplayer it is." s118 00:09:06.040 --> 00:09:13.160 And it reminds me sort of of like um actually like Midjourney, like the fact that everyone was on Discord seeing how other people were using it. s119 00:09:13.160 --> 00:09:23.480 I think it actually to your earlier question really helps with that unreasonableness or ambition where the first time you see somebody tag Claude and be like, "Hey, you know, don't just fix this bug, but like now you are responsible s120 00:09:23.480 --> 00:09:32.920 for this part of the code base and I want you to monitor this feedback channel and proactively take on tasks and then fix them and then also take like, you know, if this API changes, do that." s121 00:09:32.920 --> 00:09:34.080 Like I saw somebody do that. s122 00:09:34.080 --> 00:09:36.760 I was like, "Oh, wait, I've I've totally underutilizing this thing. s123 00:09:36.760 --> 00:09:39.440 I've just been using it as like a glorified Claude code and slack. s124 00:09:39.440 --> 00:09:43.920 Like that's definitely a totally like sort of new version of it, right? s125 00:09:43.920 --> 00:09:50.840 And then more advanced version is really trying to start thinking of it as a teammate that is actually sort of holds context, has memory, and can be proactive. s126 00:09:50.840 --> 00:09:53.840 And that's just really changed how we operate internally. s127 00:09:53.840 --> 00:10:01.560 It's much more like this multiplayer async proactive way than it is a you know, most people often their own CLIs. s128 00:10:01.560 --> 00:10:04.080 Are you bottlenecked by code review and get? s129 00:10:04.080 --> 00:10:08.280 Obviously, there is code review, but someone usually still looks at it. s130 00:10:08.280 --> 00:10:11.040 Is there a world in which you just merge it in? s131 00:10:11.040 --> 00:10:13.360 Yeah, we're it's a really good question. s132 00:10:13.360 --> 00:10:19.440 We are definitely still bottlenecked on reviews, especially for things that are like touching some architecture pieces. s133 00:10:19.440 --> 00:10:24.440 And it's actually more subtle than just being bottlenecked on review, cuz that's you know, okay, we can carve out time differently. s134 00:10:24.440 --> 00:10:30.360 It's like bottlenecked on human ability to even like fully conceptualize what we're doing. s135 00:10:30.360 --> 00:10:39.640 So, one of the reasons we built Claude code artifacts that we shipped a couple weeks ago was partially for that, which is you would send somebody a PR, and then they'd be like, I don't know, man. s136 00:10:39.640 --> 00:10:41.280 This is like 2,000 lines of code. s137 00:10:41.280 --> 00:10:43.600 Like, it looks like code to me. s138 00:10:43.600 --> 00:10:47.600 And what we started doing instead is sharing much more like, here's a Claude code artifact. s139 00:10:47.600 --> 00:10:49.200 Like, here's the explanation. s140 00:10:49.200 --> 00:10:51.320 Here's the intention of the the change. s141 00:10:51.320 --> 00:10:52.720 Here's the trade-offs that were made. s142 00:10:52.720 --> 00:11:02.480 And like, I think that's going to be much more be the trend by which we communicate, which is the code is ultimately, you know, verifiable using some things, but actually like discussing s143 00:11:02.480 --> 00:11:08.160 intent and trade-offs, and then measuring in production I think that at least the direction of travel we've we've gone. s144 00:11:08.160 --> 00:11:12.080 And I don't review when I get a pull request, I wish I could say I reviewed every line of code. s145 00:11:12.080 --> 00:11:12.880 I definitely do not. s146 00:11:12.880 --> 00:11:17.320 I like actually talk to Claude about the the code and say, all right, like, these are the questions that I would have. s147 00:11:17.320 --> 00:11:18.360 Can you go investigate it? s148 00:11:18.360 --> 00:11:21.560 So, it is kind of Claude-powered code review, but still human-driven. s149 00:11:21.560 --> 00:11:23.000 And and for the really important ones. s150 00:11:23.000 --> 00:11:30.120 And for the ones that are like cosmetic visual changes, it's much more like look like we'll fix forward if we need to fix forward, you know. s151 00:11:30.120 --> 00:11:30.960 Yeah, totally. s152 00:11:30.960 --> 00:11:34.000 I think a lot of people are here are trying to figure that out, too. s153 00:11:34.000 --> 00:11:38.400 I wanted to talk also a little bit about Anthropic Labs in general. s154 00:11:38.400 --> 00:11:44.520 Nilay Patel, who you've probably met before, loves to ask ask the question like draw the org chart. s155 00:11:44.520 --> 00:11:47.640 Like how like people, you know, you ship your org chart. s156 00:11:47.640 --> 00:11:50.320 Like I think it's important like everyone knows cloud code. s157 00:11:50.320 --> 00:11:52.320 Now you've got tags. s158 00:11:52.320 --> 00:11:54.480 Um How are you structuring the labs? s159 00:11:54.480 --> 00:11:55.720 Yeah, it's a good question. s160 00:11:55.720 --> 00:12:05.240 Because what we were trying to wrestle with was you want sort of people to be supported like, you know, I think the death of the engineering manager discipline has been greatly exaggerated. s161 00:12:05.240 --> 00:12:11.560 Like I think there's still a lot of coaching and interpersonal pieces and personal development that I think is still really, really important. s162 00:12:11.560 --> 00:12:20.560 But especially in a labs type group where like our whole cadence is two-week reviews where every project goes up for we call it persevere or pivot. s163 00:12:20.560 --> 00:12:28.040 So basically every project is up for review and either it's time to, you know, keep going, persevering, or you know, it's time to pivot it or even shut down. s164 00:12:28.040 --> 00:12:35.839 And, you know, we've shut down projects basically every single one of those cycles and it's like the more you do it, the less it's just like, "Oh no, my project is shut down. s165 00:12:35.839 --> 00:12:36.320 I failed." s166 00:12:36.320 --> 00:12:44.280 It's like, "No, that is definitely the intention of the labs team is to prototype quickly, try to ship internally, maybe get it to early access, and if it doesn't work, wind it down." s167 00:12:44.280 --> 00:12:54.000 But because of that kind of like rapid iteration, it means that if you align the org chart too much to the individual projects, you're going to end up like re-orging every two weeks, which would be a total nightmare. s168 00:12:54.000 --> 00:13:01.760 And so we've actually ended up with this interesting setup where like the the pod or the team that is working on a given we call them bets within labs, s169 00:13:01.760 --> 00:13:09.000 definitely just draws upon like all right, somebody from product, somebody from the eng team, you know, I'll jump in when it's a product I'm particularly interested in. s170 00:13:09.000 --> 00:13:11.480 I'll come in and work together with the team on it. s171 00:13:11.480 --> 00:13:12.960 And that's the unit for that time. s172 00:13:12.960 --> 00:13:16.440 And there is the concept of a bet lead or a directly responsible individual. s173 00:13:16.440 --> 00:13:23.600 But the interesting thing is that they don't manage usually any of the other people, which kind of breaks the that kind of previous way in which a lot of these things were done. s174 00:13:23.600 --> 00:13:28.640 But I think it leaves it leaves us to be really flexible when you say, "Okay, actually this project is not going to work out. s175 00:13:28.640 --> 00:13:31.040 Let's disband and keep going and it's not a big deal." s176 00:13:31.040 --> 00:13:38.640 And the engine manager is much more playing the like make sure every individual is assigned to the thing that they're most excited about and that they're working in the best way possible. s177 00:13:38.640 --> 00:13:42.120 Now, what we do sort of solidify is when there's a product that has like legs. s178 00:13:42.120 --> 00:13:51.240 Like Cloud Design for example, started in this sort of ad hoc sort of grouped way and then now that like we've shipped it, it's gotten traction, we've done like a big second release s179 00:13:51.240 --> 00:13:51.720 in June. s180 00:13:51.720 --> 00:13:57.280 Like it's becoming like we've hired people for that specific team and it has more of a of a structure. s181 00:13:57.280 --> 00:14:01.000 So, it's like loose until it gets solidified down the line. s182 00:14:01.000 --> 00:14:02.360 What's the future of Cloud Design? s183 00:14:02.360 --> 00:14:07.520 I think a lot of people are very interested in It's one of your biggest launches this year. s184 00:14:07.520 --> 00:14:08.520 Where does this go? s185 00:14:08.520 --> 00:14:16.320 I think for me, I mean the things that are holding back Cloud Design from being even better is better interaction with our other surfaces. s186 00:14:16.320 --> 00:14:21.120 So, you know, I was designing something or I was talking to to Cloud Code the other day. s187 00:14:21.120 --> 00:14:27.720 I'm like, I want a really much more seamless like what I'm talking about the design for it, you know, and then after design back to that. s188 00:14:27.720 --> 00:14:32.120 I think in general it's I mean this goes back again to kind of unconstraining Cloud. s189 00:14:32.120 --> 00:14:35.800 Like the fact that our surfaces don't talk to each other as well as they could. s190 00:14:35.800 --> 00:14:39.160 I think really holds back a lot of interesting ideas around what we could do. s191 00:14:39.160 --> 00:14:43.160 So, I think that's one like kind of major area that we're looking at. s192 00:14:43.160 --> 00:14:49.520 And then the other one is people like the lines between a Cloud Design and an app get blurry and blurrier over time. s193 00:14:49.520 --> 00:14:57.640 Like I've seen people Of course there's no like persistence but build like fully functional like even games which is definitely not what we designed Cloud Design for but you can do it. s194 00:14:57.640 --> 00:15:00.200 It's just HTML and JavaScript. s195 00:15:00.200 --> 00:15:13.680 So, blurring those lines even further and thinking through like what is the path from a like fully featured design that looks really well to really good to something that is maybe more like a artifact where you're actually able to go and you know, persist data and share it with others and build from there. s196 00:15:13.680 --> 00:15:16.839 So, I think that those lines get really interesting over time, too. s197 00:15:16.839 --> 00:15:18.080 Yeah. s198 00:15:18.080 --> 00:15:20.440 A big part of design is having taste. s199 00:15:20.440 --> 00:15:26.800 I actually asked Fable what Fable wants to ask you and this this this is what Fable came up with. s200 00:15:26.800 --> 00:15:35.200 You deleted almost all of Bourbon to get to Instagram which is like you had a whole you know solo mode whatever thing and you went to Instagram. s201 00:15:35.200 --> 00:15:38.680 What would you delete in AI or more spicy what would you delete in Claude? s202 00:15:38.680 --> 00:15:41.440 Oh, I like the spice. s203 00:15:41.440 --> 00:15:47.680 I think I mean we have it's interesting we have a one of our slack channels is like project unship which is like what is in the product right now. s204 00:15:47.680 --> 00:15:49.680 It's I mean this is hard at Instagram. s205 00:15:49.680 --> 00:16:02.320 The Instagram we what things that had like four to five percent usage you're like oh that's really not very many but then you have like 20 features that each have four to five percent usage is like the classic Microsoft Word problem of like s206 00:16:02.320 --> 00:16:06.160 everybody uses some disjoint subset of the of the functionality. s207 00:16:06.160 --> 00:16:07.800 So that that's always the challenge. s208 00:16:07.800 --> 00:16:11.600 Now I think we're younger product so hopefully we have less of those things. s209 00:16:11.600 --> 00:16:23.080 Like we unship styles I think recently where it was like used by a small percentage of people and was not really AGI filled in a lot of ways it was like very sort of prescriptive in the way that it worked and skills very much better s210 00:16:23.080 --> 00:16:24.240 applications and then like that. s211 00:16:24.240 --> 00:16:33.920 So I think you have to be willing to take the primitives of like one generation of AI and like unship them or at least like supplement them or supplant them with the next one as well. s212 00:16:33.920 --> 00:16:43.080 I think the biggest thing is I look at it and I've been spending some time like outside labs on some of this is like man like we're asking people to make like code versus co-work versus s213 00:16:43.080 --> 00:16:52.440 like chat distinctions and like one they don't interoperate well and they can't delegate to each other and two I think the average person off the street could not explain to you why those are all different. s214 00:16:52.440 --> 00:16:59.760 So I think deleting some of the product complexity within our our code or our product I think is a a thing that would would serve well. s215 00:16:59.760 --> 00:17:03.200 Also because then Claude can do what it needs to do and and do well. s216 00:17:03.200 --> 00:17:12.240 Like there's nothing more frustrating than having a co-work session where you're like great I've mapped out exactly what I want you to build and then be like can you please like create a paragraph that I can paste into Claude code? s217 00:17:12.240 --> 00:17:17.400 Like that is some 2020 you know kind of workflow there that really shouldn't exist anymore. s218 00:17:17.400 --> 00:17:17.920 Yeah. s219 00:17:17.920 --> 00:17:24.199 Um I think drawing lines on what you don't want to do and also sort of leaving room for others is interesting. s220 00:17:24.199 --> 00:17:29.960 Um a lot of people today is like the startups day for AI E or obviously very sympathetically aligned to startups. s221 00:17:29.960 --> 00:17:37.680 Uh but there's some anxiety in the room because tomorrow's Anthropic could wake up and publish some markdown files that destroy my industry. s222 00:17:37.680 --> 00:17:38.492 Um so s223 00:17:38.492 --> 00:17:39.080 [laughter] s224 00:17:39.080 --> 00:17:41.520 uh why should we not all just give up and join Anthropic? s225 00:17:41.520 --> 00:17:44.600 Like why bother starting any other company? s226 00:17:44.600 --> 00:17:53.480 Um I I mean actually joined one of the main reasons I joined Anthropic was because I saw how much this was like, you know, the models weren't that good at coding up but they were getting there. s227 00:17:53.480 --> 00:17:57.240 Like how much it would unlock like whole like next generation of startups. s228 00:17:57.240 --> 00:18:04.760 Not because it was going to solve their ideation or their taste, but because like it would make experimentation way simpler and and would get you to move faster. s229 00:18:04.760 --> 00:18:06.280 And I still like really believe that. s230 00:18:06.280 --> 00:18:12.040 And I mean it's the reality of, you know, uh And we saw this with like Instagram. s231 00:18:12.040 --> 00:18:16.240 Like we would get questions from investors like, well, what happens when Google launches a photos product? s232 00:18:16.240 --> 00:18:24.040 It's like Google's going to launch a very googly photos product and it's going to have to be bound by the integrations that they already have and it's going to be like it's going to play to their strengths. s233 00:18:24.040 --> 00:18:25.160 And I think that is going to be true. s234 00:18:25.160 --> 00:18:35.360 And I'm not like giving advice on how to compete with Anthropic, I guess in a way, but like it's actually not because we're also a platform which is like there's so much I think room to be like laser obsessed with your particular s235 00:18:35.360 --> 00:18:41.680 vertical or your industry or group of people that you know really well in a way that like none of the labs are ever going to get to that level of s236 00:18:41.680 --> 00:18:47.240 uh of understanding and like therefore get that kind of adoption and user love and and build that out. s237 00:18:47.240 --> 00:18:57.080 Now, it's definitely harder in the age where like the models can just do a lot and so there's, you know, some of these things can be like skillified and like maybe don't need their own dedicated product. s238 00:18:57.080 --> 00:18:58.800 But I think it's like the hard stuff is still hard. s239 00:18:58.800 --> 00:19:04.720 It's like understanding the needs of people, like figuring out how you're going to reach them, uh listening to them and iterating on them really quickly. s240 00:19:04.720 --> 00:19:15.720 Like it is still the case that like a group of four or five people obsessed with a problem is going to move faster than those same people at any other kind of organization that are like, you know, subject just to the complexity. s241 00:19:15.720 --> 00:19:19.920 I just mentioned the like the fact that we have, you know, a lot of different products that kind of interoperate. s242 00:19:19.920 --> 00:19:22.080 Like that's a interesting constraint that we have to work through. s243 00:19:22.080 --> 00:19:23.600 It's an advantage in other ways, right? s244 00:19:23.600 --> 00:19:33.960 So, yeah, I'm still like very long and bullish on startups and um it's just it tapers over the fact that like writing code was never the like the limiting part. s245 00:19:33.960 --> 00:19:38.680 You know, maybe it was on the timeline perspective, but it was never like the thing that was going to like make or break a startup. s246 00:19:38.680 --> 00:19:41.160 It's really that space and user understanding. s247 00:19:41.160 --> 00:19:41.600 Yeah. s248 00:19:41.600 --> 00:19:43.080 Uh domain knowledge. s249 00:19:43.080 --> 00:19:43.640 Yeah. s250 00:19:43.640 --> 00:19:47.080 Uh today is also our day for vertical AI. s251 00:19:47.080 --> 00:19:53.320 Uh one of our uh returning speakers and top speakers, Chris Lovejoy, uh was always talking about vertical AI. s252 00:19:53.320 --> 00:19:55.280 He was in from interior in the healthcare space. s253 00:19:55.280 --> 00:20:01.840 And then recently I was I invited him back and turned out he you guys just hired him for your uh healthcare efforts. s254 00:20:01.840 --> 00:20:04.760 Um we also our next big one is also finance. s255 00:20:04.760 --> 00:20:06.680 You know, we have a yeah, a finance track. s256 00:20:06.680 --> 00:20:09.440 You guys just had a huge finance event in New York City. s257 00:20:09.440 --> 00:20:13.080 Um and where our next uh AI is is sort of finance focused. s258 00:20:13.080 --> 00:20:13.880 What are you seeing there? s259 00:20:13.880 --> 00:20:16.200 Any you know, any potential uh for Claude? s260 00:20:16.200 --> 00:20:18.160 Obviously a lot of Excel Excel spreadsheets. s261 00:20:18.160 --> 00:20:18.520 Yeah. s262 00:20:18.520 --> 00:20:20.480 No, I think that there's there's a lot in there, too. s263 00:20:20.480 --> 00:20:27.480 And that's like an area where uh you could see the model get clearly better at it like sort of generation to generation. s264 00:20:27.480 --> 00:20:36.640 And there's, you know, there's some good sort of vertical specific uh uh finance startups that have like done their own um evals, which has also been interesting to to track. s265 00:20:36.640 --> 00:20:43.720 And it's not like we're like sort of playing to the eval, but it is a useful sort of barometer around like is this actually getting better um at these finance use cases. s266 00:20:43.720 --> 00:20:54.120 I think the interesting blend that's going to happen um is this mix of, again, the model having the flexibility to like dive in and create just-in-time analyses or dashboards or workflows s267 00:20:54.120 --> 00:20:59.960 with like some sense of like what is the not immutable, but at least like verified sort of set of data. s268 00:20:59.960 --> 00:21:09.520 And so like uh set having all of that be totally free form, I think is a recipe for confusion and is like not what most companies in the financial services space want. s269 00:21:09.520 --> 00:21:24.760 So, finding that right uh sort of cut line where you have verifiability and audit logging and and sort of data provenance here, but not in a way that constrains the kinds of applications that you can build on top, I think is a lot of the art that we're seeing in that space as well. s270 00:21:24.760 --> 00:21:29.760 Um and I think, you know, if you solve it well, you can you can get the best of both worlds. s271 00:21:29.760 --> 00:21:38.520 The hard part is a lot of the systems that were built to do the verifiability audibility like are kind of almost by design not super flexible in terms of agentic workloads on top. s272 00:21:38.520 --> 00:21:41.320 So, I think there's opportunity at both sides of the stack there. s273 00:21:41.320 --> 00:21:41.880 Yeah. s274 00:21:41.880 --> 00:21:45.800 Um I think I I also agree we'll be exploring that in in New York. s275 00:21:45.800 --> 00:21:52.240 Um the last thing I want to end on is on mental health, which we don't talk about enough in technical conference conferences. s276 00:21:52.240 --> 00:21:54.240 Um you've seen a lot of hyper growth. s277 00:21:54.240 --> 00:21:58.800 People are just always refreshing their timelines and it's exhausting. s278 00:21:58.800 --> 00:22:03.240 Um how do you advise people who are working 996 to avoid burnout? s279 00:22:03.240 --> 00:22:03.600 Yeah. s280 00:22:03.600 --> 00:22:05.040 I mean, I think this is a hard one. s281 00:22:05.040 --> 00:22:16.280 I mean, and it is, I'm sure you are experiencing this cuz you're all working in this industry like it is you know, multiples more intense and things move much more quickly at an Instagram s282 00:22:16.280 --> 00:22:22.960 like our the two things that we were thinking about was like, what is Apple going to announce at WWDC and is it going to like totally mess us up or boost us, right? s283 00:22:22.960 --> 00:22:24.520 So, that's like once a year. s284 00:22:24.520 --> 00:22:27.840 Um or, you know, maybe a competitor launches every three or four months, right? s285 00:22:27.840 --> 00:22:29.800 And uh it is definitely not that. s286 00:22:29.800 --> 00:22:33.800 It's a topic we we do a when we do do our weekly all hands. s287 00:22:33.800 --> 00:22:45.840 Usually on Wednesdays and we have a slide that's like the week in AI at Pinterest is and it's only Wednesday and like and inevitably like some competitor has shipped a new model and like there's been a like new product and maybe there's some interesting thing happening s288 00:22:45.840 --> 00:22:47.280 um uh on the regulation side. s289 00:22:47.280 --> 00:22:48.960 Like if things are moving really, really quickly. s290 00:22:48.960 --> 00:23:01.320 I think the way I try to stay at least relatively sane, um one is like actually carving time off and I think the topic co-founders do a good job of like saying like, look, like burnout if you you out, like you're kind of done. s291 00:23:01.320 --> 00:23:06.280 I've seen it happen unfortunately to people I'm really close to and then it takes a long time to recover from that. s292 00:23:06.280 --> 00:23:11.760 Um so actually encouraging people like there's no job that is so important that you can't be offline for a couple of days. s293 00:23:11.760 --> 00:23:14.560 Um so I think that's like a big key like piece in there. s294 00:23:14.560 --> 00:23:16.476 So like let's s295 00:23:16.476 --> 00:23:16.720 [applause] s296 00:23:16.720 --> 00:23:18.280 strongly believe. s297 00:23:18.280 --> 00:23:25.440 Um And if it is you're probably doing something wrong and you talk to somebody who could be a mentor to figure out how you can unblock that. s298 00:23:25.440 --> 00:23:34.520 Um and then I think the other one as well is like I love sports and like uh this is the notion of like you're never as good as like your best game and you're never as bad as your worst game. s299 00:23:34.520 --> 00:23:35.880 I think that's also really true. s300 00:23:35.880 --> 00:23:39.920 Like I know like in AI there's like the you know it's so over we're so back thing. s301 00:23:39.920 --> 00:23:47.040 Like that like if you internalize that that cycle is always going to be at play in some way, you realize like it's never that bad. s302 00:23:47.040 --> 00:23:59.560 Like Ben Horowitz's book is the hard thing about hard things has this chapter on like we're effed it's over and like that feeling as a startup that probably many of you have had at startups where you're like oh I can't believe this thing happened like we're never going to like recover from this. s303 00:23:59.560 --> 00:24:02.000 I definitely we definitely had an Instagram a couple of times. s304 00:24:02.000 --> 00:24:07.040 And then you get through it and like it's that like def like defines the company when you can actually go through that. s305 00:24:07.040 --> 00:24:14.160 I try to remind myself and the team here even with an Entropic which is like like this is a is a fast-moving but is also a long game. s306 00:24:14.160 --> 00:24:20.280 And it's like we're never it's never just about today's model launch and reaction or this product launch and everything else. s307 00:24:20.280 --> 00:24:31.400 Like you're playing and you're building and you just have to trust that you're building like the team and culture that is going to get through those things and have that sense of perspective even if perspective is saying like look 3 months ago we were in a similar position. s308 00:24:31.400 --> 00:24:43.120 Maybe it's not a year it's just a matter of months but it's still like zooming out and not taking things not letting your internal sort of like sense of self and success be so driven by the day-to-day. s309 00:24:43.120 --> 00:24:43.520 Yeah. s310 00:24:43.520 --> 00:24:50.520 It has anyone any coach or mentor said something to you that you repeat to yourself that gets you through the hard tough times? s311 00:24:50.520 --> 00:24:59.920 Um I think the biggest one was this like sense of like if you're feeling something it's really often the case that other people on the team are feeling it too. s312 00:24:59.920 --> 00:25:10.960 So this is like advice I got from my my coach around just being like like just verbalizing emotions like even saying like hey I'm feeling really stressed out about this or yeah I'm really sad that we are shutting down this labs initiative. s313 00:25:10.960 --> 00:25:23.160 I literally had this meeting a couple months ago where I was working really hard on something and I kicked off the meeting like I'll kick it off like I'm really sad like I'm frustrated like I wish this thing had worked out and I think that holds the space for other people to be like yeah I'm pissed off s314 00:25:23.160 --> 00:25:36.200 too or like I'm sad too and like I think giving that advice around like not I think if you can get yourself to be open and vulnerable it often like lets other people verbalize that and then you can from there you can be like great what are we going to do about it like you know it's much s315 00:25:36.200 --> 00:25:37.720 easier to start from that place. s316 00:25:37.720 --> 00:25:47.560 Yeah we actually kicked off AIE with a session from Carol Robbins who runs touchy feely at Stanford and I can't think of a better way to end than encouraging people to talk about their feelings s317 00:25:47.560 --> 00:25:49.480 manage their mental health and keep shipping. s318 00:25:49.480 --> 00:25:49.880 Yeah. s319 00:25:49.880 --> 00:25:50.520 Thanks so much Mike. s320 00:25:50.520 --> 00:25:52.920 Thanks for having me.