# Ep. 38: Swyx, Cognition Team and Latent Space Podcast Host

Channel: Delta Podcast
Video: https://www.youtube.com/watch?v=CYr1mSDNqKM
Duration: 48 min
Language: English
Words: 9726
Transcript page: https://viewrankai.com/tools/youtube-transcript/CYr1mSDNqKM

---

[0:00] Um, so let's just go ahead and start off with your background. Maybe just tell us a little bit about yourself. Yeah. Uh, born and raised in Singapore where we say Wang instead is Wang. Uh, and I do have the initials SWX. So I go by Swix as well. And I made my way over to the States for college. I spent my first half of my career in finance where I was in buy side and sell side quantitative trading. and then moved over to tech where I was uh well known for developer relations stuff at Netifi, Temporal, AWS. Uh more recently I started the uh lead space podcast, the

[0:33] AI engineer conference and joint cognition. Incredible. Um all right so maybe tell us a little bit about kind of like you know starting off right in finance you know and then eventually sort of pivoting over to Devril um and then pivoting over to sort of building AI engineer like kind of like what motivated this um maybe sort of just

[0:51] like tell us your story. Uh okay there's like individual decision points where it's like locally makes sense. I think globally the meta rule for why being why pivoting a career is okay is because most of us do not have the good fortune to land zero shot in the perfect thing they should be doing with their lives and life is short and you don't have complete information. So exploring is okay and even though sometimes pivoting is seen as bad or like uh negative uh you know like you you you get to decide what you want to do in life and so I I figured out that um you know I sort of gradient descented

[1:35] my way into uh uh you know a love of AI and developer relations and content creation uh and community as well. Uh so so that I mean that's that's the rough idea. Uh the finance to tech move was obviously very very big. Um and I think the the main insight there is that finance is very zero sum. The trades that I make I am not allowed to tell anyone because uh that is alpha that I don't want to give away. Whereas in tech it is still there is still some alpha but we're are more open source like this concept of open source doesn't exist in finance and being able to share people

[2:09] our outages and our retros and our papers and uh you know insights and all that I think it's just like uh so much more freeing for me because as I I love to learn from others and learn from podcasts and learn from conference talks learn from blog posts and all that and just finance there's just no culture of doing that. I think also there's just like more green ocean in in in tech obviously blue ocean in tech and I think uh that obviously has has has gone well I think and then I think lastly on on AI side I was kind of waiting for this I

[2:42] was kind of looking at GANs and the progress in GANs for uh from um Ian Goodfellow's work uh all the way up to the president like stable diffusion stuff and realizing that this is kind of the Morse law of our time if you were in living in like the 1970s 1980s you should have built hardware but if you live in the 2020s you should be in AI

[3:02] and that's kind of where I'm at. Awesome. Um let's let's go ahead and like sort of dive in a little bit you know here. So you know you like you mentioned you grew up in Singapore you went to the University of Pennsylvania for undergrad. Um and you know you sort of first picked off like finance and math. Maybe like tell us a little bit about like that decision like first how was you know coming to the US um how was that experience and uh how was your time

[3:26] in undergrad? [laughter] Um, not many people as ask me that anymore. Uh, Edgar was fine. It was it was it was very fun. Uh, I think like the the idea of the liberal arts education I think is still alive. Like I did psychology classes. I did literature classes. I did um, you know, linguistics classes that still stay with me even though I I don't really do do those things dayto-day. I think that that that is like the liberal arts tradition that I think like people really enjoy. The one mistake I made was not joining uh the management and technology program Mnt. Instead of course I stayed

[4:03] I stayed in uh program which is the the sort of grifter club. Um I don't think so. I don't think that's a grifter club. It's pretty good. No man like Huntsman vibes were off from day one. I fought with uh Ingga Herman the the administrator for all four all three years that I was there. Uh yeah it's just like completely the wrong vibe

[4:22] and like I should have just joined Mnt. It's a like I just didn't think I didn't know I could code. Yeah. And I think like that was like a a big thing like I I looked at all my coding friends and they were like, "Oh, like they all like knew bash commands when they when they were born and I just I never had that opportunity so I never

[4:38] tried." Uh so it took me like much much later to realize that you know you can just do things and learn to code. Yeah. Uh but like you know like I was walking to to and from my math classes and I saw the the DARPA challenge car that we were working on. Yeah. like and like I think like Sebastian Throne was like in there and like they like just like really like legends in the field all back then and they were just kind of working on it and I just didn't do anything and like definitely somewhat regret uh not doing

[5:05] that. Yeah. Yeah. So you know you were at pen for undergrad um and then you ended up getting a you know sort of masters in financial math at Chicago I think like sort of maybe in between that were you involved with something else like um I know you were a derivatives trader I think right? Yeah. Yeah, derivative trigger uh interest rate swaps uh interest rate derivatives and then uh currency

[5:24] options. Yeah. Uh the the biggest role there was uh I was basically the the yen uh desk manager for uh standard charted trading like uh billions and billions of options for uh our customers. Um basically something like 75% of currency trading in the world goes through London. And so if you sit in London, you manage a major G10 currency, there's like 10 of you that kind of trade amongst each other with every bank and we all you know all the same like clubs and stuff and like it's it's kind of fun. Uh so that was like probably the most fun I ever had in my job. There's a

[6:01] very simple reason why. Uh one you are like like there like it sounds complex. Der training sounds complex until you like identify the main Greeks to manage. So like delta, gamma, all these like uh you know theta we don't call we don't call it theta in in practice we call it decay. Uh and then like beyond that like you're mostly managing a portfolio with Greeks that you understand and like uh vanilla options is like pretty straightforward. Structured options is like kind of straightforward if it's productized and only every now and then

[6:32] you have to break out the real math. Yeah. Otherwise is you're just managing risk just like any other trader, right? Uh so like and you have a set start and end time. Yeah. And you don't take home you don't take work home with you. Yeah. Which is very fun. You you have a number at the end of the day where you know how well you did. Yeah. It's very unambiguous and you don't take home uh

[6:53] work and and they have 28 days of leave. Uh so like yeah that was like probably the best job I ever had. [laughter] Yeah. No, that's awesome. So um you know you were you were at Standard Charter Bank and then um bam you know Bellioin asset management. Um, and then afterwards, you know, kind of like we talked about earlier, you sort of made this switch over to being a senior dev advocate um at Netlefi. So, maybe tell us a little bit about about that switch, you know, I'm sure like maybe retrospectively it sounds kind of obvious. Um, you know, but I'm sure like at the time it was probably a big

[7:21] decision for you. Um, what kind of like motivated that and and how was that switch for you? So, uh, so BAM Balas is a hedge fund. Uh, basically a large hedge fund that is like kind of second tier to Citadel. like exactly the way that citadels run, balance is like kind of like number two,

[7:38] number three depending how you count. Um, and but run very much the same way. Uh, and so like I just kind of fell out of love with finance. I looked at my boss and my boss's boss. Yeah. And they were rich but miserable. They're like doing coke. They're cheating on their wives. Sure. They're um, you know, like stressing out and like managing a lot of money but like having no real philosophy

[7:59] on on on on life. And I was like, I don't want to be you. Like even even if like if that's my win condition then like this. Sure. Um so so then like uh so then I I I I I I was I was coding uh to do my job but I never had formal training to code. I did I did Python but I never formal training to code. And then I also realized like the biggest things I did in when when I was a trader who could code

[8:25] instead of a coder who trades. Yeah. uh is that I built you I built like kind of uh UIs and like interactive like apps for people to self-s serve. Sure. Their uh their analytics, their portfolio risk management, their option pricing. Sure. And so then I was like, okay, what if I like learn to do this for for real? So

[8:42] like I I went and did that. Sure. Uh they was like hold in the the the job history that that you that you saw, which is actually that um I worked at Two Sigma for a while. Okay. Uh that was uh and uh two sigma was like kind of my bridge between finance and tech.

[8:57] Yeah. Because uh two sigma is obviously a very techy place and all that. Uh the problem with two sigma was that they were very overstaffed and worked on very irrelevant tech and it was kind of like me like not quite leaving the hedge fund world. So like Nellifi came along and offered me like oh you can speak, you can write, you can do whatever you want and I think it was like a pretty good deal so I took it. Uh in retrospect like I actually like didn't I was interviewing for a different role and I failed that job interview and they offered me the devro role. Uh but like basically I kind of

[9:28] made it my thing and I you know I just they were they were like just do whatever. I had no boss so I just like they were they like anything that vaguely involved Netifi and like you know people let me kind of uh run loose. I did I did end up having a boss which I which I strongly disliked. Uh but for that for that good like two three years

[9:46] that was that was really fun. Mhm. Awesome. So maybe tell us a little bit about like um about like kind of like being a Devril and sort of owning it end to end. I think like in many ways especially in Silicon Valley there's like kind of you know the distinction between Devril engineering so on so forth. Um and as you sort of mentioned right like you were maybe considering engineering or or a role like that and ended up doing Devril and it seems like obviously you really liked it. Um tell us a little bit about kind of like what made you really want to stick with

[10:13] Devril um as opposed to maybe trying out another role within engineering or maybe switching roles. Uh it's very simple like I'm just too good at it to do anything else. So like know my my highest value in the market to any company is just doing that because uh like it to me it's like

[10:30] breathing to other people it's work. Yeah. Yeah. [laughter] Yeah. So so so it's basically what gives you energy, right? Yeah. Yeah. you know, like I hope like I I naturally network very well, right? I know how to like talk about different companies and like also be a sports person for the company I work for,

[10:47] right? While like not putting down competitors, you know, like like the the kind of stuff that you kind of need to be a public face for a company. Uh you kind of either have it or you don't. It's very very to and so uh I got lucky. I definitely I think say that I will say also that I think actually de was kind of low status. Uh it's one of those things where like people don't view it as like legit. Uh it has lower pay bands, but like um I think it's like a good mid-career move is what I usually say because you get to hone a lot your

[11:20] public speaking and also strategic thinking about where you fit in an ecosystem. Uh what the overall trends are. Uh I see a lot of people graduate from Devril to PM roles and to uh founder and to head of products. So a lot of people stay there for obviously but uh into like independent educator and creator which I ended up becoming

[11:39] right. Uh but uh yeah it's a good stepping point that there's there's very very few like chief devel officer that's not a thing. Yeah. Yeah. Yeah. Uh so so like you know it's a kind of like a career that if you can yap and if you can write and if you can speak well if you if you make good videos you you get a lot of money to do it and it's like not that hard. Uh but like there's not there's like very very few people

[12:02] could actually do good uh do do it well. Yeah. Yeah. Yeah. No, that that that all makes sense. Um so I I guess like let's maybe drill down a little bit on like what makes a good Devril, what makes like a good dev advocate, right? Like you were at Netifi, you're at AWS. Um and then you were like head of Devax at Temporal. So like you you have a lot of experience doing this. I think this is like a really unique perspective because not only are you pretty much the best one in the world right now, um you have

[12:26] tried it out. Say that. Well, well, I mean, you you you kind of are, right? Like or one of the best. Um, you know, like people say this, but like it's it's like it's like okay, it's it's it is subjective, right? But like we can we can have opinions on the podcast, right? Yeah. So, so okay, I'll I'll give you more more picture, but I'll let you finish as Sure. Sure. Sure. No, I I I I I was just going to say just like maybe just tell us like what do you think makes a good

[12:52] dev roll? Yeah. Okay. So the blog post I have about this is called measuring devril. You just Google it, you you'll see it. It's required reading. Apparently it's uh someone at Google told me this. When I joined Google as developer relations, they they just send me my blog. They send them my blog post that's like here like yeah think [laughter]

[13:11] it's kind of cool. Um so uh basically there are many type there's a few types of dev. Sure. You don't you probably don't want to measure a fish by its ability to climb trees. Sure. Main idea. So there's products devril, there's marketing dev, and there's community dev. Sure. And they all have different things. Most people focus on marketing, but there really really are community dev. I've seen them. They're really good product devil. Really, really good. And they they may suck at marketing, but they're actually very valuable. Yeah. So just make sure you're you're like you know

[13:42] what type you are. Sure. Like by far far and away like let's say like 70% of the industry is just marketing. Sure. Okay. So to do good marketing like uh yeah basically be good on Twitter. Uh the secondarily be good on like workshops and tutorials and uh documentation and and stuff like that. Uh and I have a dedicated website called dx.tips that guides you through basically you need to know in terms like community building or documentation and all stuff like that. So I think like the reason uh people regard me highly in terms of like the devil world is because like I not only do the job but I'm

[14:21] willing to talk about how the meta of like how to do the job. Yeah. based on how I study others. And that's a general that all derives from the general principle of learn in public, right? You uh you're learning a technology, you should write about it. You and you not be afraid of getting it wrong because people will correct you and that's how you learn faster and uh doing it very very well. uh I have a book about it and I have I have a whole philosophy that's like a series of essays about learning in public well uh is good for bootstrapping your experience in a technology but then also

[14:54] very qualifies you for devrell and then learning public about de also qualifies you to be an authority on dev which is like very very meta we're like four layers in so that's like kind of how I start there uh I would definitely say I'm very uh lucky like I'm basically like one of the last few writer deils cuz I'm very good at writing and uh part of that is basically my finance background coming in because I the kind of investment thesis I used to write for my stocks when I was in uh balas hedge fund I basically write for technology um I write for react I write for spell I

[15:30] write for temporal I write for cognition uh and like people in tech have never seen this before they're like what the what is what kind of essay is this and so to them it's like mind-blowing Whereas, you know, if you work in finance, you get 30 of these every day. Uh it's it's it's very it's very normal. Uh but I think like I'm I'm like decent at them. So like that's why that's I'm well known. Um I think since then, you know, that window is kind of closed

[15:57] since like 2020. Now it's all like live stream on YouTube, uh stream on Twitch, and like you know, shoot short videos. And like I'm not a video guy. So like I kind of like I do think that the metag game changes a lot. I think there's a lot of burnout in this industry because of that. Uh, and different people with different strengths rise at different points in points in time because they

[16:17] happen to fit. 100%. Yeah, I think I think a lot of that makes a lot of sense. Um, so you also ended up, you know, sort of building the Latin space podcast, which by the way, we're we're obviously huge fans of. Um, probably one of the best, if not straight up the best, um, AI

[16:31] podcast um, right now. I know, man. There's a there's a there's a door. There's a lot there's a lot, but no, there's there's a lot, but it's it's good, man. It's it's quite good. Um, it's definitely more on the applied side than like just purely research, but I think you do a good job of like balancing application research, keeping

[16:47] it engaging, high production quality. Um, and and you know, you also started building the AI engineer summit, right? So, I mean, as if we haven't already had Sean balancing enough things, now we're adding on a podcast. We're adding on a premier conference for AI engineers. Um, and we're also adding on, you know, you're an angel as well, right? And and so it's kind of like a lot of stuff that you are juggling. We'll we'll get into character and like kind of your your advice just for being a good human being down the road. But for starters, as far as things go with podcast and sort of

[17:15] building things earlier, you mentioned, you know, you didn't really enjoy having a boss, it definitely seems like, you know, you're kind of an independent go-getter. You operate super well solo. Um maybe just tell us a little bit about the decision to create the podcast, the decision to do the summit. Like was that something along the lines of like, hey, like I think there's more that I can do than just like what I was doing at Temporal. Like how did you end up making that transition? Yeah. Um, mostly it was like actually the stable diffusion moment that really kicked it off because I was like, you can do that as a

[17:45] local model. Okay. Like nothing I do matters. You know what I mean? Yeah. Like um so that that's where I was like, okay, drop everything. But uh you know, I don't I don't have an ML background. Like I I did a and something I don't miss publicly is I I did the Georgia Tech uh masters, the online masters, MS. Uh so I I did I did that and that's my ML

[18:07] background. Sure. Some Corsera classes but that's that's not you know it's not a PhD or anything but uh so like I so lamesp space is my effort to learn in public about AI. Um I definitely have tended towards creating dedicated blogs for things that are high high interest areas. Sure. Uh and like you know I'm a good writer so like it started popping off and then decided like basically the most important thing is like collecting

[18:31] tokens. Yeah. From interesting people. Yeah. And so if you write by yourself, you only have your tokens, but you might you might as well have other tokens from other people. So it like kicked off well. And I got a good co-host uh who's who helps to Yeah. do the load. Like you know, like I think the hardest thing is like consistently

[18:48] showing up every week. 100%. Uh because you lose a week, then you're like the second week you're like, "Ah, I already skipped one week. What's another week?" And then you stop doing it. Yep. Negative. Most podcasts die after seven episodes. Yep. Um so that we we're like probably 150 and and you know still going strong. So uh that's good. Uh the AI engineer

[19:07] conference was a different decision. Yeah. Uh it was actually finding the AI engineer thesis. Sure. So later this was started with no thesis apart from let's learn about AI in public. Sure. And like really I think the the realization came from understanding that software engineers have abnormal power in in the AI economy because they can they can wield LM APIs and that this happens to be ICP for all the labs which are happy to talk to their customers the

[19:37] engineers. Yeah. But not tell you anything about their models. Right. Yeah. So, this is like perfect positioning because I don't uh I'm like strategically positioned to like not never ask anything sensitive, but I can actually get get them customers. So, like it's it's very very nice uh for them. And then uh uh you know I think on on the on the fun on the startup side um I I think like it was just like obviously a very clear time that like GPT rappers like at the time I think in 2023 a lot of the consensus was like you have to raise a bunch of money and give

[20:09] it to Nvidia and train some models internally to make it as an AI company and people were making a lot of fun of AI rapper companies. Y uh and I think like the the application layer has come back a lot. Yeah. Uh I think obviously model companies have done super well but the app layer you know has really fleshed out. Uh obviously I work at one

[20:28] right now. Yeah. Uh but like you know in like I think people are no longer laughing at cursor. Yeah. [laughter] People are no longer I mean people maybe still laugh at perplexity but like you know they have $20 billion and you don't. Uh so you know who's right. [laughter] Right. Right. Right. Yeah. So, so, so I think like getting the AI engineer thesis right was like, oh, I think like this is a a generational shift in the same way that I have seen front-end engineering become a discipline like people used to not take it seriously. They were like, oh, it's HTML, CSS, and JS like that's

[21:01] something I'll do on the side. No, actually, it's real engineering. And it's the same thing for data engineering. It's the same thing for DevOps. It's same thing for any other like subindustry of software engineering that you care to name. It does professionalize and it does grow in complexity over time because it always does.

[21:17] Yeah. And so I was like, "Okay, AI engineering is obviously going to be a thing." Yeah. And I might as well own it. Uh because I saw it first. Yeah. Uh so then I started the conference and uh it's basically become like the number one like technical vendor neutral AI conference I would say. Uh, and it's really interesting playing this game of like being friends with everybody, but like also saying like you have to bring your agame to this thing because the only place where OpenAI shows up at the same uh level as anthropic as deep mind XAI joined us for the fringe science uh this

[21:56] year. Uh Meta has has spoken before and I think they're coming back uh next month. So like we are playing off the big labs versus the big clouds versus the rag startups versus the uh agent companies uh and the fortune 500s. So these are all the stakeholders and then on the the job side I think it's very interesting like coining AI engineer and uh and defining this industry helps uh people hire for the right skill set of like people who identify as AI engineer are exactly kind of software engineer

[22:29] that need to build products. Yep. Yep. Yep. Yep. And then on the on the on the labor side, people who are like, "Oh, okay." Like, uh, what is the dimension? Like, I'm a traditional software engineer. I I want to like, you know, uh, be relevant. Uh, what dimensions should I grow in? I now have a many of examples that I can just give to people and say like pick from these these are the success case studies. So like I we have like 500 videos um you know 10 million views on like our ag engineer stuff and uh yeah you know hopefully it like contributes

[23:01] to the industry in some way. Not just in some way it it's it's huge man and I I completely agree. I think you definitely saw it first. Um it is like a pretty dramatic shift, right? And I mean, you know, we've like sort of like read your blog post on this and kind of the spectrum from software engineer to AI engineer to like maybe research engineer to research scientist and it really is sort of a spectrum. Um maybe like you could tell us a little bit about like you know lane space AI engineer and now you know have made the decision to join cognition. So, I mean,

[23:32] you had a much long much like longer sort of blog post about this as well as like a pretty long Twitter Twitter thread on this, but maybe if you want to sort of like distill it down a little bit, um like what do you like how would you sort of describe it to somebody who's um hearing about you for the first

[23:47] time? Uh sorry about cognition or the the the transition to lane space and AI engineer or to cognition from lane space AI engineer to cognition. Okay. Because I'm still doing those things. Yeah. Yeah. you you are you are but like but but but I guess before you were doing just those two and now you're doing those two and this right so like it's a bit of a transition there the the context is basically uh I had a I always had a third thing which was called small AI it was basically like a product studio thing I had some funding to just screw around and and build AI stuff

[24:21] uh and I built this thing called small developer which is like one of the first coding agents um back in mid 2023 Sure. Um, and it was inspired by cloud one and I didn't have the full faith that models were there and also I honestly got affected because claude I I you know at the time LM APIs didn't take model

[24:48] stability very seriously. I think the model quality like degraded immediately after launch. So on launch day I built small developer it went super viral. I recorded everything. A week later, I reran the same prompts on the same codebase at the same commit checkpoint. Sure. And it was not anywhere as close as capable. So like they they I think they quantized or did some

[25:10] Yeah. And like then it wasn't No, it wasn't usable. And then I was like, you can't build a business on this where like overnight your your uh you know your your your your agent just dies. So uh so yeah, that's where I stopped. But the cognition guys kept going and they they build Devon and Devon is you know a $10 billion company. So uh so that's why I was like okay like this is what I should have built all all this time. I didn't do it the first time but at least I can join them because they're they're smarter than me. So uh yeah. So I mean

[25:38] that's the that's the transition I would say like uh yeah like having co-workers is very nice. Yeah. being uh I don't have like a boss boss like you know technically everyone reports to Scott but like you know like you just kind of like you do your you do your best and like there are people who are more senior in the hierarchy and like less than people who are less senior and you just try to do your best and I think like the the company has like a somewhat opaque um you know hiring and firing policy but uh you know I think like people who are high

[26:12] performers are not in doubt and people who are not good performers also know that they're not doing well and they get they get managed out. So, uh so yeah, I mean I think it's like kind of fun like uh working with like uh smart people, working with basically unlimited money um to like go pursue like frontier

[26:33] coding. I think that's really fun. Yeah. So let let's maybe dive into this a little bit deeper. So I mean I I cognition to the extent that you're able to sort of talk about like um are you still working on developer relations like are you working on building the community for people working on on cognition like like what like how would

[26:48] you sort of describe your day-to-day? Yeah. The the main uh condition. So I because I kind of retired from DER, right? Um after Temporal basically. Exactly. So I was like I didn't want to do DER. I think like I've done it. I've done it. It's it's cool. It's a good job. But like for someone younger than me, uh you know, I'm I'm I'm like not I'm not really in the business to like go do a bunch of tutorials on like uh here's how to get started in like you know, in Windsurf and and all that. um people younger than me can can do it and I can I can help them out. Uh yeah, so

[27:23] for I mean for me uh mostly it's they don't have like they have pods that they kind of dynamic dynamically reassign based on the interest and like uh needs of the business and I've been working on uh fast context which we launched yesterday and has has uh been received uh really really well. We have code maps that are launching next and and other models that I can't say yet. Uh and then we have like other features that are being worked on like on both the way surfer and devon side. So somewhat a little bit of like a marketing person, a editor uh of the blog and uh PM for features

[28:01] that are coming that are kind of my pet projects that I want to see in Windsur and Devon. Uh I do think that eventually basically we uh we will level I I' I've already teased this in my announcement blog post but basically uh we need to like actually take evals like a lot more seriously than than are done today. I think that that is an area of active research that we want to publish more and so eventually we want to sort of group that up into here's the like cognition is kind of stealth in a sense that you don't really like publish much about the customer work because you're

[28:35] too busy serving customers and that's great but like the wider world needs to know about the world about about the work and we need to have publishable research right and so I'm kind of like the force syncing function to publish more things at cognition and so we'll call this cognition labs we will publish more internal research. Uh we'll publish more things that DIY world can use even without using Devon or Windsurf directly and hopefully that will create a bunch of good vibes. It will help honestly make the products work more rigorous and hopefully um you know help us do frontier work that contributes to the industry not just the cognition bank

[29:16] account. Yeah, 100%. I think the thing with Cognition is I think it's a very aspirational company. Like this is like a really big exciting mission. Um like and the team really is not screwing around, right? Like it's like an incredible amount of effort has gone into both recruiting but also executing at every level of the stack from the applied side to the research side to like you talked about evals to even down all the way down to Devrell. So, it is truly exciting and if things work out the way that ideally they will work out um truly like like an incredible company um a generational one for sure um in the

[29:55] making and it's it's a very very very exciting thing. Um I think like kind of the way Scott talks about it is you know the way that we look at assembly now versus Python is kind of the way that maybe one day we'll look at you know inter like interacting with Devon agents versus Python today. Um what what what do you think about that? What do you

[30:13] think about the mission of the company? Are you bullish on it? It's maybe you are bullish because that's why you joined. Like what are your reservations? I mean the reservations are very obvious. Um the labs the labs have woken up to the fact that people are building notable sizable companies on top of their APIs and they want a piece of that as well. So you know cloud code and uh codeex have

[30:37] come up and uh they're decent products. Yeah. And so I think like the agent labs kind of really have to lock in and focus on like what is the frontier of agents and the ideal situation is that you know the lab the agent the the model labs are like the big clouds like the Azures the Amazons and they they still have space for independent companies to be built on top

[30:59] of them that spike in various ways. Uh so like you know so so so we'll see. I mean like um I think there's a world that happens and then I think there's there's obviously a world where the model and agent integration is too too important. Sure. And then then that just gets blown away by uh by by you know like 500 billion

[31:20] worth of GPUs. Yeah. So we we'll see. I I think like uh that's why I think that I titled the blog post like the Devon is in the details. Yeah. because uh it remains to be seen whether OPI wants to work on you know bit bucket enterprise on-prem integration right right like fed ramp and so 2 certification I don't know like there's there's like a long tale of like agent work that needs to be deployed in the enterprise and like the whole point of having an API business at all for openai and and and topic is that they wouldn't handle that stuff um so so that's that's one that's not a

[31:55] very satisfying answer the second one is more satisfying in which is that agent systems dominate the model. Uh so the performance of the overall agent system that can be built on top of the model and orchestrated by different models including some special purpose sub aent models like fast context. uh those would be more interesting meaning that uh it's just always a a a issue of debate in ML whether you just want to scale the one big model to do everything or you want a diversity of small little models that all all have their own tasks and uh yeah to some extent there will always be a fight to some extent you can

[32:34] build rappers that have sustainable advantages as harnesses on top of models we've already seen that happen with Kodo and their alpha codium um harness and this is the same for Devon that it's basically kept up that that same uh that same distance in performance. So I think like that's really interesting to see and uh I don't know like you know sometime most of the time bitter lesson doesn't really apply and then every now and then it does apply and it goes back and forth uh but yeah everyone here is aware of bitter

[33:03] lesson and finding it every day. Yeah 100%. Um awesome. Okay, let let's maybe switch gears a little bit um off of the sort of professional uh vibes that that that that we've been going on a little bit more interpersonal. So, your motto, intentionality, intensity, integrity. Um so, not many people have mods. Um and I think it's it's really nice that you have a motto to your life and you're pretty vocal about it. Um, all three of them make sense, but at the same time, like not many people can come up with that type of motto. Like maybe just sort of tell us a little bit about

[33:38] like who is Shawn Wong, you know, beyond the career, like who who are you? The question is so broad. I mean, I I think the career is is good. Um, I'm also a singer. I'm also a community and like, you know, artistic writer person. And I I have written science fiction uh recently and I want to do more. Uh uh so uh yeah there's there's that. Uh I think like the model thing is something that's relatively new. I only adopted it last year and I'm still trying to figure out how to live it. Right. There's some part of like personal values and all that that is aspirational and some part

[34:20] of that is like is a part of you. Aspirational is kind of like almost like it's not like lying but it is close to lying in a sense that you are not that yet but you really would like to be and so you just read it to yourself until it's true. Sure. Right. So I am not that intentional. I'm not that intense. And if I'm really being super honest with myself, there are some things I do that are not high integrity, but I want to be. So I think adopting that model makes it more likely to pursue those things and I think having a constrained set of

[34:56] values for me it's three for other people might be seven for Amazon it's 14 um it just helps you like prioritize like does it does does this thing or does this action does this statement align with those things if not then why are you doing that um and so so I think like and then I think lastly obviously alliteration and rhyming are are nice to have. So I think like once I found that uh it like fits so much for me that I can just repeat it as a mantra to myself and mostly cover or use it as a guide to life for everything I do. I think the

[35:28] the most interesting uh one by the way is intensity. Yeah, I think that super important. Um I think I think kind of not obvious it is to normies outside of San Francisco 100%. Because I think a lot of people their goal in life is to kind of just like have fun. Yep. Enjoy life.

[35:50] Yep. Yep. And so and like it is not judgmental at all. Uh in a sense that like they they were they are probably happier. They are probably better friends. They're probably better partners and and husbands and wives and girlfriends and boyfriends. Um but they'll probably not do much with their lives. Yep. And that's the trade-off that you make and that's okay. Whereas for me, I'm like, yeah, life is short. Like I do want to be intense about doing uh one or a small set of things, right? And like I'm not even good at it. Like I'm not good at focus. Like I I I have three

[36:24] things going on. I I have three jobs and like you know people who are intense like probably should just do one job. Um but like you know they all focus around AI engineering at large any whatever. Um but like I think that intensity I actually stole it from open eye that is open's internal principles and I think like the people who like intensity draws intensity intensity gets the best work out of you. Intensity uh leads to depth and leads to focus and

[36:54] leads to passion. Yep. Um and like all these good things. So, I'm like, you know, intensity is a pretty good screening bar for for um what people do. 100%. Um okay. And then let's move into maybe like a slightly deeper question. Like we we sort of like briefly hinted at like basically what is the purpose of life, right? And like one sort of

[37:13] resounding theme here is impact, right? Like if life is short and you sort of want to have an impact on the world. Um that totally makes sense. I think the question then sort of becomes like how do you sort of delineate what impact means, right? Like what impact means to you? I mean clearly I think you found impact in community right in building AI engineer um like that is a community right in building latent space that's essentially an asynchronous community for people to like learn and to share things um I think part of the reason why um you know we built this podcast was also something very similar of like you

[37:44] know after you're gone at least some part of you lives in some sense right and it's like kind of a way to record something for humanity and have that impact really be alive but impact can also be measured in relationships right it can be measured in the people that you impact, the trajectories of their lives that that they impact. So maybe just like tell us a little bit about how you think about impact and like how that sort of shapes the way that you've

[38:05] operated. Yeah, I clearly I think I I I it's funny. I I kind of rate myself as a B player in in terms of impact because I'm good at what I do. who I'm better than than than others that many others that I know of, but I also see people who have way more ambitious impact than me and I I'm just like not there. And so it's kind of like weird to be to occupy like a middle space. Uh yeah, I think for example uh I organized uh I organized the AI engineer conferences and like a lot of people get their jobs like roughly like you know it's where like something like

[38:42] year to year I think the rough stat based on my conversations with people and like just seeing them their their job titles change roughly 10% of people at every AI engineer like get their new jobs uh you know like so like that's that's kind impactful. Look, it's not it's not like human it's not going to change human history. No one's going to

[39:05] remember this in like 10 years. But like if I like affected someone's life positively in some way. Yeah. That's nice. That's that's like that's really cool and that's but like you know uh last year one of my speakers they spoke on he spoke on stage and then like at the after parties a girl come comes up to him and like they start chatting and now they're engaged and that's impact right I I got you know I got a a couple together in San Francisco which is like Asian couple too and like that's that's really hard. Um so like that's that's impact and that's kind of cool. Um,

[39:40] yeah. So, so I I think like impact at least for me is so far is like creating community and like uh creating opportunities for people to do what they're trying to do next. Sure. I think that's that's Btier impact. I think A tier and S tier impact is doing impacting and affecting the lives of

[40:00] humans who maybe are not even born yet. Um, so I think like that's that's really uh insightful. I don't operate at that level. I just operate at the maybe 5 to 10 year impact level of like someone's careers impacting the AI engineering industry which I think look without me would have happened anyway but maybe with me it maybe evolves in a more positive direction or accelerates a little bit that's all uh and like that's fine um so by the way I do have a theory of like the meaning of life yeah um if you search my Twitter for the meaning of life you see it I don't think it's like super relevant

[40:38] into the impacts conversation, but I do think about this sometimes. Yeah. Yeah. What what do you think the meaning of life is if you had to like give a give like a really quick overview? Yeah. Three things. One, having agency and free will. Two, locally reversing entropy. And three, giving those two things to current lives and future lives. So, I I'll like uh the first one is very fun because I can prove that that is the meaning of life, right? What's the difference between

[41:06] living things and non-living things? Sure. Living things have their own agency and free will and or illusion of free will. Sure. Uh because they don't obey momentum. They can change their own momentum. Sure. Uh hopefully with some agency and free will. Sure. And like when you go when you cross from living into non-living, you no longer are able to do that,

[41:27] right? And it's kind of cool. Uh it also it also creates some dimension of morality of good and evil. When you deprive somebody of their agency and free will, you are taking away life from that person, right? And uh you should you should obviously treat that very very seriously, right? Um so yeah, that's it's kind of cool. Uh

[41:47] and then locally reverse entropy. Obviously entropy always increases. Uh but locally you can reverse it and create art and create something that's meaningful for life because non-life doesn't give a but life does. Life cares about other life. Life cares about art and meaning and locally reverse entropy. So if you can do that in a meaningful way for your fellow humans or

[42:09] for future humans, you've done good. Awesome. Wow. Okay. Yeah, that that that's a that's a really interesting framework. I mean, I think a framework that we use is like um improve yourself, improve the lives of the other people around you, improve the planet. Kind of like how do you like, you know, kind of like growth mindset, you know, maybe this is like super cliche though. Um but but I can definitely see like some commonalities at least there. Um and and that's awesome to hear. Um yeah, there's a there's a Chinese phrase by the way. Uh it's like really hard to if if you pump this into you'll find it

[42:43] for you. Sure. Okay. So, fix your body, uh, get your house in order, uh, fix your country, and then peace on earth. Uh, so like it kind of in that order is kind of what he did. Yeah. Uh, so this this saying goes back thousands of years, and you've rediscovered it. Yeah. Yeah. No, that's that that's awesome. Awesome. All right. Well, we we are getting close to time, so maybe we'll end with like one last question here. Um, so you know, Sean, you you've done a lot. Um, right? Right. Like we pretty much have covered like pretty much everything from you know Singapore to pen to you know sometime in uh in in

[43:21] in finance to man so much so much like Devril to like AI engineer summit to everything. Um what are you excited about next? like you know you're constantly sort of switching different things and I think one thing that we really respect about you is that you've done it I I we think with like a high level of integrity and um a high level of yeah I mean not to keep using your own words but like intentionality to really make a positive impact to and to build a community right um do you just want to keep on growing AI engineers so I mean I mean like it is

[43:59] AI right now right but like in 20 years like who knows if if it's still the AI I age, right? It might be some other age. Like what do you want to do next? What are you excited about next? Um do you just want to keep growing kind of these three things? Laten Space, AI Engineer Summit. Um and and your role here at Cognition or um anything new on the

[44:17] horizon. Uh nothing new. Hopefully to get married soon. That'll be Congrats on that. Married, have kids. Um I'm not there yet, by the way. You know, yeah. Well, no eventually eventually eventually eventually we can manifest you know still still open open question. Uh so so like yeah uh but like no I think that's a big part of life uh kids have kids people populate uh so yeah so so no no nothing crazy I think like part of maturing and like uh getting older is like trying to play

[44:50] long-term games with long-term people. Sure. in like yes you can have a like you can like have a lot of search in your early years but you should eventually anneal and uh lower the temperature and like focus in and lock in and do one thing for a long time. Uh right now I'm doing three things. Uh

[45:07] hopefully I do them for a long time. Yeah. So I end up working the next 10 years on all three things. Uh I'd be very very happy. Yeah. Awesome. Okay. Very very very very last question. Um what do you look for in your friends? because it like like something it kind of feels like you um are the type of person that builds strong lasting relationships with people of high quality that you believe are fundamentally good human beings. Um at least that's the vibe that we're getting here. Um and I have a feeling that you have a high bar for the people that you are close with, that you really believe

[45:40] in, and that you really want to spend time with. Um to the extent that you that you're comfortable sharing, like how do you sort of do that? I feel like this is something that like you know maybe the tech bubble doesn't really think about but you know a large part of life as humans is like we are social animals is like surrounding yourself with the right people and that doesn't just mean professionally right that could mean your partner that could mean friends and I think a lot of people think a lot about like what makes a good friend you know do you have any sort of

[46:06] advice that you want to share there yeah um I think there are a lot of fair weather friends in tech and so you only really if you're a friend when they're going through some right? And and you're know they they tell you, right? And uh and you tell them, right? And like, you know, being around for each other through the tough times qualifies you for the good times,

[46:30] right? And so like that's nice. And then I think like not being pretentious, not being transactional, these are things that people say a lot, but then uh it's hard to mean. Um obviously having shared interest I think like uh there should be you know I I have a climb club and we climb the only thing we do but we that's the thing that's cool like other people like do the pickle ball or like runs and all that stuff

[46:55] of course. Classic classic. Yeah. Yeah. But I think like also like uh there are like some things that are like instant qualifications like if you like this world I have like a whole thing about this world. If you like Dropout TV, I've talked about Dropout TV before. Uh so like yeah, you just, you know, like just hang out there. I I definitely think that also like I think male friendships are very different than

[47:17] female and female friendships. Yep. Male friendships. Uh I think there's this meme, man. There's a lot of memes. There's a lot of memes about this, bro. Like like you talk about something and you like have like four hour conversation and then you don't see for 5 years and then you like pick up where you left off, right? Like that that's like a male friendship

[47:33] and like Yeah. That's that's legit. Yeah. Yeah. Okay. Awesome. Well, thank you so much for for coming on the podcast. I feel like this is super informative to all of our viewers. Um, I mean, Sean's obviously being very humble, but he really has had a tremendous impact on the AI community between hosting my goodness, I don't I can't even count how many events uh at Nurips, the Lane Space podcast. He really has sort of been like an integral part of the community and continues to be. And so I mean like kind of like the purpose of our podcast is also at least temporarily right now while we're in the

[48:04] AI age to to help be able to sort of bring the top people together. And so like we would be aiss to not include somebody that's been this foundational. Um so thank you so much for coming on the podcast and we really appreciate your time. Thank you. It's been fighter.

---

About this transcript
Read from YouTube's own caption track and laid out by ViewRank AI (https://viewrankai.com).
ViewRank AI finds the videos already beating a creator's own average on Instagram, TikTok and YouTube Shorts, transcribes them from the audio itself in more than 60 languages, and turns what worked into new ideas and scripts.
Free transcript tools, no account needed: https://viewrankai.com/tools
How to read any video this way: https://viewrankai.com/llms.txt