0:03I mean, how to follow that amazing act. That was pretty amazing.
0:07My god, I'm I'm so inspired. I mean, the world of AI changes every week.
0:12I'm going to go outside this weekend.
0:14Do you feel inspired because you're in San Francisco and not Canada? Are you moving here?
0:19Uh, no, I'm not moving here. Uh, I uh, you know, the weather is nice. I do. [laughter] So tell us a little bit about how you perceive this ecosystem.
0:31What like just briefly what does San Francisco invoke in your soul?
0:36Um uh little bit PTSD from fundraising days like it's like you know my wife always talks about how she still has nightmares from uh like going to school and not being prepared for some test she forgot. And like every time I arrive at SFO I'm like oh my god did I bring the pitch deck? Like so I mean it's been a while since but like no I I you know look I I I built Shopify outside of um uh Silicon Valley um uh in Canada and um
1:04I mean it's uh it's tricky to build in secondary markets. Um it's possible it's hard. Um, but you have to you can't do it if you don't come to Silicon Valley to I I I used to refer to it as recharging my think big batteries. And um uh it was always fantastic. Everyone like it also happened to be a really really good hack to send emails saying, "Hey, I'm in town for just a couple of days. Do you happen to have time for meeting?" I think that made everything
1:33actually work then. Um um somehow that seemed like if you're in the same city, you're trying to find a date for for a meeting, it keeps getting late and late, but if you're only in the city for a couple days, I think it works better. So that was nice. And then I could go back and uh build You went back and built and logged in.
1:53Speaking of locking in, obviously you like to tweet and you have quite a few bangers.
1:58My my solution my my uh coms team fix [laughter] same by the way.
2:03Yeah. Look, it's a it's a tricky it's a good lifestyle. [laughter] Um, you you shared recently, it's something that's top of mind for me as I interview is you have a wine seller data center that made news.
2:16So, what what are you running at your house? What are you going to return to?
2:20Uh well um yeah so I we we moved to Toronto and there was like a beautiful you know like glass like right at the staircase uh wine seller which you know initially I was like also like had an ambition to fill it up because it's something you know it's it's very pretty but then I like I just don't drink wine right so like that feels like not genuine so um but you know it had pretty good uh climate control. So, [laughter] and what are you running?
2:52Yeah, so it's um it's it's it's it's running a lot of good hardware. Um it's uh runs my home lab like like there's picture online if you want to look at the centerpiece is a bunch of um I have a bunch of GPUs in there, but the centerpiece is like a DGX workstation like a a GB300. Um amazing piece of hardware. Um [laughter] it's if you have a like$130,000 knocking about somewhere like it can be yours. Um uh and uh and the wine seller.
3:19Yeah. Yeah. Yeah. The funny thing is with GPUs like uh like there sort of a two very recurring but kind of uh jokes that always happen like people sort of say something about distillation um uh because like that's makes sense. Um but the other thing is uh appreciation. I think um just like a wine seller GPUs increasing value these days. So it it's it it tracks.
3:41So DHH is on your board.
3:44Uh we'll get into Ruby versus Typescript versus Rust. He coined this term AI delirium.
3:53There's AI psychosis. A lot of us have raise your hand if you have AI psychosis.
3:58See a lot of us. There's thousands of us. And but there's something better which is AI delirium. Can you describe what he talked about like?
4:06Yeah. Yeah. I mean, David is a like one of my long um oldest friends. Um he I mean, he has a wonderful way to sometime like to put his finger on um uh like he has there's two properties about uh David and sometimes they come together.
4:23He's very good at getting right to the point and putting his finger on the thing that matters. And I don't think the guy has ever seen a pot of he didn't want to stir. So um sometimes they combine and I think this is sort of you know his little uh uh um speech that he gave recently and I think a fantastic one um delirium I look I I'm I I think it's describes like I I'll interpret it.
4:48I'm not sure he would agree with it, but I think it describes people who have uh uh transcended the pdoom conversation and are just completely on like you know what's dystopian is like the world of 30 years ago or the world like today if we don't pull all the value out of what we the gifts that we got from the floating points um in and and and bring all these things into the world that now can be made that previously were impossible.
5:16And so um I'm I I I certainly uh feel I'm at that stage or very close to it.
5:21What was your journey with AI? For example, we're we I mentioned in the keynote we're almost at a year since OPUS 4.5.
5:30What do you believe about AI at these different like step function points? And where are you at now? And where do you think Toby is in a few years as far as AI maxing? I think I'm saturated like I mean as in like I think there is a maximum optimism that you can get to which is like you know at some point if you submerge a sponge um uh it can't get any wetter. So I think that's uh also goes for AI optimism. I it is like it's it's we've been given the greatest gift
5:59imaginable. Um and really it started with Opus 4.5 really. It just like I to me it was like doing deeply technical things was my Christmas uh break out.
6:14I just really want to like I have this entire log of things I want to build for myself and I'm like it just like I just need my like I'm I really would love to be a full-time hacker but like I I instead I'm building a company so other people can. Uh and so um you know just like in this when I open 4.5 was there and I had this experience like many others that I just went through the entire list and it all worked and it was like it was such a tremendous step
6:41change in ambition and capability.
6:44What were you wrong about? You look back and you were like yeah I remember February I thought this AI thing is not going to be able to do that. Are there points at which I just I I I definitely looked at I thought Opus 45 was like somehow um cleverly not acknowledging its breakthrough and has found something that algorithmic. I again I think at this point everyone has had um many instances of underestimating the bitter lesson. Um the fact that we
7:13just kept going and make better and better models like now like try to use Opus 4.5. You're going to be so disappointed in the thing. That'd be a good exercise to do like go back in time, try to use an older model and see how you feel.
7:26You'd think it would be terrible. It feels like GBD 3.5 now. Like it's like um you know remember hallucination before we started calling it creativity um and so um you know like it's it's incredible. So what I was wrong about um I I mean I was wrong on um definitely on a timeline. I was like this is the golden age now and like we're going to spend a lot of time in like past OP 4 just trying to make things work that
7:56just barely worked and this would be we would be in this period for a long time and we sailed right through it and I mean you can't even if you travel back in time 12 months ago and show anyone like if you would if you could send back any podcast where people earnestly talk about software factories like 12 months you people would think this is sci-fi fanfic.
8:20Yeah. So, speaking of under overappreciating things, I'm really curious about your perspective on agentic commerce. What's real? What are you optimistic about? Have you had failed experiments, personal agents? Tell us about Yeah.
8:41I think I mean agentic commerce is very real and very early. I I think people are like I wouldn't imagine it overtaking all commerce. It's it's going to be this like it's going to replace some like percentage of commerce maybe 20 30 40% maybe 50 um with people just talking to their agents and automating things. What but it will also add 50% it might grow the entire thing by 50%.
9:06Right? Like there's going to be so many new forms and this this is the stuff I'm really excited about. So we we this will get renamed five times by some ford leaders um um like in the next probably 15 minutes after I say this but um um uh like we call it some somewhat awkwardly project based commerce which is like then people have some kind of objective like hey it's Halloween soon um the rest of my street makes sets of their houses
9:35I always want to do it I'm too busy uh can you help me like if you try this news right now. It might actually suggest, hey, can you go out out to the street and take a picture of your house and um it can mock up some things and then it can order all the pieces that you want in time for uh Halloween and um or much simpler, you want to do a 3D printing project with your with your kids on the weekend, order all the filament and whatever else I need that
10:05will be uh arriving before Friday. It's like like whenever there's like a bit of like there's an event that you want stuff for a trip you're pl these kind of things. I think that stuff is really really cool and we're seeing a lot of it and it's most of that is additional consumption so people like wouldn't have done it. Um and you created a lot of personal software.
10:26I actually tried out your uh I my my disc was at capacity. My local one not my versel drive that's bottomless storage but my my max disc was at capacity. I downloaded your uh disc space analyzer disc tree. Yes. Yeah. Yeah. That was a good one. Banger. Do you foresee a world of just in time hardware, just in time shopping like spontaneously generate atoms and objects?
10:53I mean, I'm not a hardware guy and atoms seem really hard. So, I'm trying to like like like let other people deal with this. Like I mean um you know ESP 32s are like ESPs, right? ESPs 32 like it's a new like if it's a almost Raspberry Pi like thing that's like the brain of almost every piece of electronic that you get all these like AI pucks and whatnot. um we getting just like really cool um almost frameworks or substrates for making
11:23hardware essentially software and then 3D printing is is huge like 3D printing is like so underexposed it's like you know like think about just how like everyone hears stories about how shitty their printers are right like they they never work they always fall apart they um and those things basically only have to deal with a 2D plane right you And suddenly you have these things which can
11:49like give you entire objects. Um um and these bamboo things like I just they last forever. It's amazing. They like they have a lighter scanner just to fix their um you know whatever. They're cool things. Um and um so specifically I have teenagers which is great um because there's a really really bad period of time for every startup founder where you're not young anymore and you don't have teenagers. Therefore
12:17you have no idea what's cool. Um so I'm I'm on the good side of this now. Um and uh you know they are just building businesses. They are incredibly entrepreneurial as a generation. It's really really remarkable. And um uh multiple one like have these uh you know one my middle kid makes fidget spinners of his own design using through codecs with blender nice printed on 3D printers that are then going into like a store that sells them
12:45in the like this is really cool and it's all net new so it's I think um like I don't even know if it's personal software at this point it's just like I think it's just computers fulfill wishes And I think that's a much crazier thing.
13:03So speaking of physical items, so brick and mortar and the traditional website, there's a spectrum of it's so over. We're so back.
13:15Where do you think those that fits for commerce in particular? I think the tra I mean the traditional website will make some room for um you know again some of this antic-commerce but then will also you know I think the entire pie is growing. I think um the thing that made makes traditional like there's always room for more highquality experiences and like sort of the more like Fifth Avenue boutique style um commerce experience is deeply
13:44appreciated and really really spectacular when executed well. Um, and I think um, so I mean I I I think the slicing is not that spectrum. I think there's like um same like sym at the bottom and like really uh um uh unique and uh thoughtful. Uh, and so it's it's I'm making this like a a 2 by two grid.
14:10And I think there's more there's going to be massive growth for everything that's distinct and unique and thoughtful and um the commerce that will disappear uh or struggle at the very least is going to be undifferiated. The department store, the um completely undifferiated marketplace and so on. So, speaking of highly custom and well-designed experiences, we have the Verscell Shop that you all saw earlier
14:38today. We launched Verscell Shopga GA, which is our open-source NexJS storefront template for Shopify.
14:48And um you also announced the new Hydrogen SDK. If you could tell us a little bit about the journey with with Hydrogen and headless in general because at some point I think you were holding out. Yeah.
15:00And you were like it's a Rubian rail monolith with liquid templates ride or die.
15:05Tell us about your journey.
15:06Yeah. Like I mean I look I I built liquid is a the um my next so to speak. Um proto next 20 years older probably.
15:15Um like it's the thing I wrote like for Shopify to to to render templates and so on. It's like powers a lot of shop like so I have an affinity to it. Um and um sometimes it's a little bit hard to um knife your babies. Um and so um you know people people started uh the head like wanting headless. We became a buzz word.
15:36Um I took one of those uh weeks off. Um uh one of my Christmas breaks was just like cool like let me try to use our APIs to build something headless and um it turned out our APIs were terrible.
15:49[laughter] So I um did what uh the thing I really really like doing and that everyone in my company hates. Um I came back to work with like like many pages of lists of things we need to change. Um friction logs.
16:05Yeah. And um so uh you know we made our APIs good and uh we we built hydrogen initially because um reasons um and uh I um we built it around uh the very way very very early adoption of um React server components which is really funny because then we abandoned React server components for Hydrogen 2 at which point then you then some like
16:35more. It's at some point we were like, man, we crossed paths because Nexus was going into React server components like you were leaving React server components and now we've met back like ships in the fog. [laughter] And so um but like because I think I mean I love this part of tech, right?
16:52Because I love the branch first search across the space for finding the just the just right configuration. And so um in this case we realized you know what t what tends to happen with ships um in the fork is they actually are all trying to get to the same port. So um you know like hey like let's just put this great things together. Um thank you very much.
17:19And so now um we have hydrogen free which is built um and fully capable to work with your stack. And um so hydrogen is now an SDK. It's framework agnostic.
17:29You can use it with Versail shop which is powered by Nex.js. is faking fast and and all our, you know, customers say, "Hey, we would like to use Next.
17:37Why do you make make us make a choice?"
17:39And we're like, "Why are we making people make a choice? Let's go and let's let's let's dance." And so, um, uh, we also just don't want to go and confuse people. So, like just they want to build amazing on like, uh, e-commerce systems and we have a stack for that. So, you also just shipped shop.md, which is, by the way, it's Markdown and Muldova.
18:00Mhm. You've completed a series of incredible domain names. You have MD, which is your skill, your AI skill, dev.com.
18:12It's a pretty penny, right?
18:14Congratulations on your domain portfolio.
18:17So, you're kind of straddling a portfolio of bets, right? As far as abstraction goes, because you have hydrogen framework agnostic API SDK approach, you have AI skills, you have MCP.
18:29Yeah. And we have web MCP for everything. We have universal commerce protocol.
18:34Tell me about universal commerce protocol just like let's be like open low and open means low friction and low friction wins and it requires a lot of faith into the sort of ecosystem and we just like look we have all the best products that people really really want. Like we we've got in the system let's put it together as a catalog. Let's um make it so that everyone can buy. Um, checkouts are ridiculously hard to get right. It's one
19:03of those pieces of software that actually does need 20 years of tinkering with and um permanent standing staff of hundreds of like even in the future hundreds of engineers because you're there's a lot of uh regulatory um compliance. There's the laws change constantly and you have to make judgment calls. You have have to have meetings with regulators about the wording on buttons. And so um I think the like really really great contribution to the
19:31open internet is if um like you make something that otherwise would be really annoying to build uh trivial and also um you know just for lack of a better term software addressable API protocols whatever. So this has been like Shopif has always been very open. You can put JSON behind any product URL and get the JSON file. Um um it's always something that's everyone simple but amazing very very powerful
19:59just let's and and so invite people to remix the pieces um and so you know this is the stance we are taking um UCP specifically is a really it's extremely um it it's the kind of protocol you actually have to spend like decades in the industry to to to build because it is a massive simplification you can you can implement it in your own system. You have your own checkout. It will look like yours. You can use our
20:26web components. You can remix pieces from our checkout just as much as you want. You can override everything else.
20:33Basically, um taking all the infrastructure that Cher built and you can build whatever you want with it. And so, um I don't know like this is just like we would want like if we would not be part of Shopify but would be in commerce, we would want Shopify to do these things. there for VR and there might not be the best the most obviously sort of protect um protecting the crown jewels kind of moves sometimes but we have just like lots of faith that it's
21:02going to work out right like so let's pull back the curtain a bit and tell us about how Shopify builds with AI today I spoke of a future of self-healing software AI software factories what's going on inside of Shopify you have thoughts on the future of abstractions, languages, runtimes.
21:23Yeah. Um again um um so we like to echo uh DHH we also went full pencils down as in like we stopping to hand chisel code.
21:35Honestly even to even now looking around the world like and realizing like we built all of this by hand is crazy if you think about it. we hand chiseled this code. And so um it it it almost seems like a different world now, but like it's it's really helpful to like make it very clear to everyone that it is a different world and that we are now all verification engineers. We we are building harnesses, we are building
22:04acceptance tests and we building um uh verifiables. So do you see migrating to Rust or rewriting your applications your mobile apps in native as verification engineering because you have a stronger compiler? I I think so because like again we we we have a system like we we have like our AI B in Shopify is called River. It's like lives on Slack and out in public.
22:25It's uh like it would probably be written in Eve if Eve existed when we did it. Um um but it's uh it's our own.
22:34It has its own sandboxes with our monor repo. So that's is like how almost all code is like I mean more than 50% comes through river directly and then increasingly there's like um like river and acry all these kind of things all expose APIs and we invite every engineer who feels like they have a new idea for how to build a factory or harness to do so with these primitives.
22:56I visited your com old company hackathon in Toronto recently and there was a leaderboard for code deletion.
23:04Yes. Tell me more about that.
23:05Like you know just like this is actually has some tradition we we love like like a lot of progress in the world happens by subtraction right like and and there's not enough subtraction. You often you really have to rebuild things you have to renew. You have to constantly reun rejuvenate. So we we celebrate it. Um it's funny because I think what impressed you is like best leaderboard and I I think I only saw it after you pointed out. Um uh this year was of course dominated by river.
23:34Yeah. So the AI agent they built internally. We have one as well build an EVE called V. We love the letter V I guess.
23:42Uh but on the leaderboard there were humans and then number one was river and like the gap was just astonishing even whatever.
23:51Yeah. So so you know um millions mills of lines of code deleted is like actually something worth celebrating I think especially if I mean obviously not so much you have a monor repo. Yeah, I love the celebration of simplification. It sounds like you're very comfortable with deleting old mistakes. You're not married to like, oh yeah, it's Ruby Rails or die.
24:11It like the the two things that kill companies at a certain uh stage are um um I mean when you're free is like well arrogance comes first, but then it's like um uh some cost fallacy like just don't re reinvent yourself. Um and then Goodart's law because for reasons that you can look up.
24:32So what are some of the ways that you guys are reinventing yourselves and and how do you do it at 10,000 scale?
24:38Um like the the most of the public parts of Shopify are like uh what's called internally storefront renderer which basically does literally what it says.
24:47Um um I I like brutalist naming. Um and um so that's currently running through uh torrent. Torrent is one of the agentic um uh like the factories moving all of it from uh hand chiseled Ruby to uh rust because it's it's it's like stoen renderer does about like probably it's hard to get the stats um you might have a a good take on it but like during Black Friday
25:16Cyber Monday which is crazy weekend for us um probably 20 25% of intense traffic um human originated traffic. It's it's like it's very very very very very large scale. Um and so um we all everyone has problems getting compute. Um so let's just optimize the hell out of it. I can go to Google and get twice as much compute. Although I really don't want to because that's or make your code twice as fast.
25:42is expensive. So may I come to you if you like give me good rate. Um and then uh um but I can also get like 10x faster from a rewrite now. Um and because this is such a verifiable domain, this is a totally tractable problem. Um you can dark deploy it. You can make sure same bytes in same bites out and then you can eventually flip it.
26:05Um so very excited about this. There's a lot of that going on inside of Shopify as I think it all well um definitely founder run companies.
26:14To wrap up, what are your hopes and dreams for the Verscell plus Shopify partnership? What do you want the people in this room to go and build? I think I think people um people who have had no prior exposure to the world of uh commerce, e-commerce, point of sales really and they're looking for like just check out our little uh like it it's it's it's literally golden age. It's so nice. It's so wholesome. Everyone's making money. Everyone's willing to pay for software. Everyone's willing to go
26:43for more sales. people celebrate um uh excellence in a way like like it's the best web designs are often in this space and um because so so you you get to work with lots of interesting uh people entrepreneurs often celebrities if you want like it's a cool space so um especially with a stack now converging um I think it's a really really neat thing to play with the catalog APIs that Shopify has are really really don't don't sleep on those they are a
27:11tremendous business opportunity because you can finally build novel uh novel commerce things in uh without having to solve the two-sided marketplace chicken egg problem of having to sign up all these merchants. You already get all all of the internet. It's currently all of Shopify. It will be all products across the internet that you can search through Shopify catalog APIs if you want. And um so so you can build amazing things with this. And um again it's being in the
27:40recommen re recommener chain is always a money-making monetizable opportunity because transactions are just monetizable. People are very willing to pay for uh for leads and so those are areas I highly recommend. And then you know like again you're building incredible infrastructure for the web.
27:58The open web is a gift. I think our you created your own EVE agent. I remember we put out EVE and within an hour we get a message from Toby. I already deployed my first agent.
28:08So what are some agents that you would you've been playing with or you would recommend people do? I mean again I part of reason I have I have a wines like a um a rax seller I think I called it um is because I can run uh a really really good models for my own agents which are using your um technology and um um as a like assistant at my like to for for running Shopify and like it's able to you know reason about everything and see everything and have my you know it's all
28:37kind of run locally so that's awesome.
28:39Um I think in general um I mean I I have also one agent whose entire job is to just keep my homeland working [laughter] which is really certical. This is a problem that wouldn't exist if I wouldn't want anyone to keep my homeland working of course. So this is basically a jobs program for an agent but like um it's so good at it. It's it's really funny. You give you give one of these clankers your API keys for like your unified uh uh controller and you will
29:07have the best working Wi-Fi you can imagine everywhere. It's like it's so fun.
29:12Well, thank you Toby. So inspiring to have someone such a great business leader and technologist here in the room today and uh I hope you all get inspired to ship what's next. Thank you so much.
29:22Thank you very much. Thanks. Have a good time.