E182 - AI Vibe Marketing Is Here - Berner Setterwall & Tom Ström

 In Episode 182 of the AIAW Podcast, we’re joined by Berner Setterwall and Tom Ström, Co-Founders of Cogny, to explore how agentic AI is redefining marketing orchestration. Moving beyond traditional automation, they share how AI agents can execute, coordinate, and manage complex marketing workflows—from campaign execution to technical operations. We dive into the shift from SEO to Generative Engine Optimization (GEO), the role of Model Context Protocol (MCP) in enabling real action, and what it means to build autonomous, AI-driven marketing teams. From the convergence of marketing and engineering to the evolving role of human creativity in an agent-powered world, this episode offers a practical and forward-looking perspective on the future of growth. 

Follow us on youtube: https://www.youtube.com/@aiawpodcast

Transcript

Terafab And The Compute Crunch

SPEAKER_00 0:00

I think most on podcasts and things like that, but yeah.

SPEAKER_07 0:03

Yeah, exactly. Listening to Elon talk about it is is my primary source as well.

Anders Arpteg 0:08

Yeah. I'm surprised sometimes it's not that it's not like the first page news throughout the world, more or less, because it's such a big thing.

SPEAKER_07 0:18

Yeah, and it's almost also the secondary, I think. I mean, obviously the Dyson swarm is like where it's headed, and it is. It sounds completely insane. Like we've heard about it in sci-fi for so long, but like actually he plans to build it in 36 months.

Anders Arpteg 0:32

But but let's do the setup then. Yeah. No, but okay, so for one, you know, Elon is building this ship fabric uh or factory, and it's not a small thing because you know he's basically saying that today, if you just combine uh combine all the chip output that we are producing per year, it's basically 20 gigawatts. Yeah, and he says basically in in Tesla and Optimus Bot and in SpaceX and on XAI and everything, he will need uh like 50 times more, yeah, like one terawatt. So that's a terafab name. So okay, if he wants to use uh more compute than you know 50 times more than's being produced per year today. That that number in itself is just insane.

SPEAKER_00 1:16

And I mean, just the space and energy you need to run that.

Anders Arpteg 1:20

Yes, so yeah, that's the the second thing then, but also you know, we have a big um bottleneck in terms of who is really creating the compute. Yeah, of course, we have the uh European ASML company that's building the machines, the lithography machine, yeah, but then it's being you know designed by Nvidia and then um being manufactured by TSMC in Taiwan, and then no one can really compete with them um in terms of the most advanced ships, and and that creates of course a big bottleneck. So it also makes it really hard to innovate because if you have a new design, a new mask that you have designed for the ship that uh NVIDIA does, then they have to send it to uh TSMC and they manufacture something, they try it out, and they have this kind of really long loop to really do innovation months, months, yeah, if not more. So now what TerraFab is planning to do is to take that on themselves. So they um they can basically do the design themselves, they can do the manufacturing themselves, and they can do the the use, the application themselves. Meaning they can really take that loop, the iteration loop for innovation, and put it down from yeah, probably years to to weeks or something. Yeah, and no one else is anywhere close to it.

Henrik Göthberg 2:32

But but is that feasible or is it the Elon Musk claim?

SPEAKER_00 2:35

I mean, I think he'd done it with the rockets before. I mean, when he was starting SpaceX, he went around to all the rocket manufacturers and asked, I need a rocket that can do this and this and can launch and you know, relaunch and and so on. And nobody said that they could build it, so he had to build it himself. So I mean he'd done it before. Yeah, he has done it before if you look at it that way.

Anders Arpteg 2:52

I agree with that. But then even the what you mentioned with energy, you know, the most crazy thing about Terrafab, I think, is they're going to build it on the moon.

Henrik Göthberg 3:01

Yeah, that's the logical extension of the energy implication. If you want to build Terrafab, what's the energy implication? Let's put it on the moon. Yeah. Is that part of the plan? But like because it's it's a logical consequence of saying Terrafab.

SPEAKER_07 3:15

No, but he's saying like within 36 months, the cheapest place to deploy a new AI will be in space.

Henrik Göthberg 3:21

And long term, I fully I think he's correct. When if he if we have cracked how we put things in orbit, and then he cracks how he puts things cost-effectively on the moon. Yeah.

Anders Arpteg 3:34

When he's using the Artemis, you know, moon um uh project now undergoing with NASA. You know, he's going to use the starships to actually put the people on the moon. It's not going to be the NASA rocket, it's going to be Starship. So he actually gets NASA to pay for him buildings like Starship rockets. So he can then use it to build the Terrafab on the moon.

SPEAKER_07 3:56

Yeah, or it will like outcompete them on the moon. I think they plan to put a couple of people, and he can plan to put like a whole AI factory on it.

SPEAKER_00 4:03

So I mean it's such a smart and strategic move, but I mean but uh at once, I mean he's always said that he's gonna do this. Uh and it it's kind of now that you can see the clear view of it or the whole plan, but he's always said that uh this is the whole plan. And I mean, if he can build that much AI, he I think he said himself in the in the interview that there isn't gonna be a single problem for a single human being on earth that can't be solved.

Henrik Göthberg 4:31

Well, then comes another question: do we need that much AI, or will we fundamentally come up with smarter ways of you know, slimmer ways of doing things? I mean, like is the scaling just gonna continue, or are we gonna think about fundamentally different architectures where this is actually not the way we're gonna scale it? What do we think?

SPEAKER_07 4:51

I think it depends what you want to do, right? I mean, I think EU was thinking in these terms for quite some time. Like, we should draw down all of the energy used, like, minimize, try to preserve Earth as it is. But Elon Musk's plan is to consume 3% of the moon and turn it into silicon in the Dyson Swarm. And he's like frustrated that we're still like not a Kardashian one civilization yet.

Henrik Göthberg 5:17

So I need you to unpack the sci-fi reference here. Dyson swarm, what is it?

SPEAKER_07 5:23

Yeah, so so basically, I think all of this goes back to this Kardashev, who is like a Russian physicist during the Cold War. He comes up with this scale of civilization, where like Kardashev scale one, if I'm not mistaken, means that basically you capture your whole planet's all energy that it gets basically. And then scale two, you capture the whole sun, and then people start thinking how do you accomplish this? And then it basically means that you build like a solar panel around the whole sun, it's the most efficient way because you can't really go there with a boiler or anything because it will just also turn into fission so you can't extract energy without building solar cells. So that's the Dyson sphere. The swarm is just like starting to build that sphere because it's just like a Starlink style satellite swarm.

Henrik Göthberg 6:09

Storm that moves towards the sphere.

SPEAKER_07 6:11

Yeah, exactly. Like when we reach level two, I suppose we we take all the energy from the sun, and then as you said, Anders, there is like the galaxy, the it goes to the universe as well.

Anders Arpteg 6:22

Continuum, yeah.

Henrik Göthberg 6:24

For all the sci-fi nerds out there, right? But maybe if it starts so nerdy anymore. I don't know. Exactly.

SPEAKER_08 6:29

It's gonna be in the reality near you very soon.

Henrik Göthberg 6:33

Well in in a mailbox near you, so what was the in an outlet near you?

Anders Arpteg 6:37

But but it's also very you can think about the reason for Elmas to do so. Uh one is simply that it's he's he means what he's saying, meaning he wants to make human species multiplanetary, or he wants to accelerate the transition to sustainable energy, or abundance, as he now calls Tesla's mission, right? Yeah, or it's actually something else. Uh and and this is simply something he says to really get people excited about working with us because it's you know a sci-fi movie more more or less coming to reality. I mean, I think it's a super smart thing to really concentrate power now, what he's doing.

SPEAKER_00 7:12

Yeah, I mean, uh have you read the about the IPO he's gonna do on two trillion dollars. Yeah, it's actually trillion now, right? It's insane. So I mean, he's gonna I think like as an entrepreneur, like it's not only the money, it's it's also about doing something fun. I mean, if I had those kind of uh influence that he has, maybe I would build them.

SPEAKER_07 7:34

He's always already the most uh like rich person on earth, so like he can't compete with that anymore.

Henrik Göthberg 7:40

But if you read the biography uh from Elon, uh what's his name? The famous guy who wrote it, uh write it. Walter Isacson. Walter Isaacson, that's the one. I mean, like there is a pattern here that is actually not so dumb. If you put an extreme goal, like we did in the 60s, let's go to the moon. So when you put that extreme goal out there, it drives you to solve problems that you then figure out and commercialize on the way. So if you think about it, by putting up the fundamental goal of going to Mars, he had to build SpaceX. And in order to find SpaceX, he had to build Starlink. You know, so so all the time it's a little bit like if you put an extreme goal up, it gets you focused, and then you figure out what is the commercial models along the way. Yeah. And and it's not, I don't know if that's bad. I think it's pretty brilliant.

SPEAKER_00 8:33

Yeah, no, I agree. And we talked about innovation before, and I think like um during World War, for example, there's been a lot of innovation during the Apollo program. There's been a lot of innovation. So I mean, there's definitely room, I think, in humanity for people to set these goals.

Henrik Göthberg 8:48

And I mean, maybe if they're just set them, where is the innovation gonna go and come from? Then we're gonna have to, you know. I love the advertising. You know, what was the top European innovation? We put the bottle cap that it don't fall off, right? Then you don't have enough strong goals if that's the main thing you can come up with as a joke, right? Yeah, of course.

Anders Arpteg 9:08

I'm already looking forward to to the end of this podcast. Yeah, after work, go even more philosophical.

Henrik Göthberg 9:14

And uh and by the way, try those beers upstairs. Yes, the homemade beers. Oh, yeah.

Meet The Guests And Backstory

Anders Arpteg 9:18

I'd love to try those. But before that, I would love to welcome you here, uh, friends, ex-colleagues, and I think extreme AI and engineering experts, Berner uh Satervel Satterbal and Tom Strom. Welcome here. Thank you very much. And uh we uh we worked together a while back in uh Campania, and it was one of my most I think fun times, and I learned so much as well.

Henrik Göthberg 9:43

Yeah, when you did a budget according to a hockey stick, I heard that one.

Anders Arpteg 9:46

Yeah, yeah, we were young. I still remember you know the days when we had this competition with uh Netflix to win their worldwide marketing, and we won that you know, worldwide.

Henrik Göthberg 9:58

Uh that's such a extra Netflix, Salando, yeah, some pre-Kickass, some kick ass customers.

SPEAKER_00 10:05

Salando, yeah, yeah. But I think uh Netflix was a lot of fun because of course they were respected as an engineering company, yeah, and their own engineers were kind of like, oh, why are you taking in this external smart engineer? You know, we can beat them, and then we kind of had to rally our engineers saying, like, let's go. Yeah, exactly.

Anders Arpteg 10:25

Respect inside Netflix, that's pretty cool. And then we had the um the office right above um yeah, a famous nightclub.

SPEAKER_00 10:41

Yeah, it was a very good office. A lot of karaoke nights, and yeah, it was I mean Is that when your karaoke career started? No, I came up with karaoke. A DE was a big uh you know part of that.

Berner’s Path As Self-Taught Engineer

Anders Arpteg 10:53

But definitely I brought the quality down. I love it. Anyway, we are here to to hear more about uh your work and um especially with Cogni and uh your current approaches to to use AI and agentic AI for marketing and so many more things. So we'd love to dig deeper into that. But before we go into what you're currently working with, perhaps we can have a short uh background about each of you personally. Uh, if we start with you, Berner, how would you describe yourself? Who is really Berner?

SPEAKER_07 11:26

Yeah, I mean that's a complex question, of course. Uh obviously, like my professional self, I'm a self-taught coder. Uh I went to gymnasiat. Uh I was supposed to go study at Handels, was my plan. Like, I had the grades and everything, but then I was like, maybe I should just try like one company before I start studying.

SPEAKER_00 11:46

Dip my toe.

SPEAKER_07 11:47

Exactly. And and I managed to land a job at one startup, and then I went to the next, and and then I uh co-founded Campania and here I am, still uneducated. And then in my spare time, like I like to be out in nature. I drive my snowmobile, I have a bunch of dogs, I have a family, two kids, and my son we live in in Yemtland. I moved there in 2020.

Henrik Göthberg 12:10

Where do you live in Jemtland?

SPEAKER_07 12:11

Uh in between Östersund and Oren in the countryside.

Henrik Göthberg 12:14

Because I've been thinking deeply in these ideas.

SPEAKER_07 12:17

Yeah, in a small village called Alsen.

Henrik Göthberg 12:20

Alsen.

SPEAKER_07 12:20

Yeah, it's a beautiful lake. It looks almost like Switzerland because you can see like Oreskutan in the background.

Anders Arpteg 12:26

Let's talk about this later, yeah?

SPEAKER_07 12:27

Yeah, uh it's nice.

Anders Arpteg 12:30

I've learned so much from you. I I was always amazed by your both AI skills but also engineering skills. So um very impressive, I must say. Oh, thank you.

SPEAKER_00 12:39

That must be good to have a PhD saying that to you.

Henrik Göthberg 12:42

So let me just do one follow-up question. What's the trick to become a great engineer, self-taught? How did you go about that?

SPEAKER_07 12:51

I think that I was just like really, really, really interested in computers from an early age. Like, it's it's not that I was like without help as well. Like, I have my brother who's working at Ericsson, he's he's done his whole whole career there. So when when I was like uh maybe seventh, eighth grade, like I got my first you know, thrown out computer from their office, installed Linux, started playing, you know, deleted the whole file system many times. A lot of mistakes, but I also spend like a tremendous amount of time in the terminal.

Henrik Göthberg 13:24

Yeah, it's a tremendous amount of time in the terminal, TTT, right?

SPEAKER_07 13:29

Yeah, something like that. Yeah, so when close code happened, I was just like rejoicing, like in deep in my heart. Like I there's always been this chasm between like developers who spend a lot of time setting up their environment and those that do very little. I'm I'm on the little side, like I'd rather be able to work just with VI on any server that I SSH into. Um, so the whole like terminal format of cloud code was like really I love that.

Tom’s Marketing Journey And Campania

Anders Arpteg 13:56

Yeah, yeah, I can see that. Tom, how would you describe yourself?

SPEAKER_00 14:01

Uh well I'm Tom. I um you know I'm not a self-taught anything. I would say I went to school and um you know graduated the Royal Institute of Technology. I was um focusing on kind of entrepreneurship and marketing. And then I went to uh Shipstedt, where I basically you know Afton Blowlet was going through this very kind of big change, you could say, uh, where a lot of advertiser was leaving, everyone was becoming digital.

Anders Arpteg 14:34

What year was this approximately 2010?

SPEAKER_00 14:37

Yeah, something like that. Yeah, um yeah, so I stayed there for a while and and basically um helping them to buy companies and to run growth inside their portfolio of companies, and then I kind of joined uh uh campagna team. How do you meet burner at campagna?

SPEAKER_07 14:56

Oh, okay, yeah. It was, I guess, through like uh our shared friend uh Mark because you were working with him part-time on Constant, right?

SPEAKER_00 15:06

Yeah, yeah, yeah.

SPEAKER_07 15:08

And then like just uh gravity pulled all of this together into Campania kind of or momentum.

Henrik Göthberg 15:15

How big was the team when it was at the biggest? At Campania, yeah.

SPEAKER_00 15:18

Was it around 50 or something like that? 5050. And we had offices in London and Palo Alto. Palo Alto, of course. Yeah, of course.

Anders Arpteg 15:27

I mean, cool. And um, now you have uh a new company and you're working with that. Can you just describe briefly how how you got started with uh both I guess uh guess growth hackers and also cogniz?

SPEAKER_00 15:40

Yeah, I mean I think uh after Campania, we saw that an Indian American company. Um so we founded then growth hackers because we wanted to go a little bit wider because Campania was very niche that we were doing just bidding for SCM basically, or search engine marketing or Google Ads primarily. Um so we wanted to go kind of wider, work with more channels, also kind of follow the users into the web page and app. Uh and that's kind of what what uh Grosshackers is basically helping companies drive.

SPEAKER_07 16:13

There were some key moments, right? I think like we worked with Netflix, for example, helping them roll out ads across uh like they uh they were successful in the US, but they didn't have a global footprint, so they came to us and like help us run ads everywhere, pretty much. And we rolled out you know country after country, and then we launched in Sweden. There was a launch party, they came here, we were all like hey. But then uh it was also like getting the performance, getting the users cheap enough, getting the scale so that they could take this like streaming platform position that they have today. And then they had like zero Swedish TV shows on the landing page, or like the content that you have, like it's not even available here because of content restrictions. So it's like maybe optimize this page and we can do a better job optimizing your ads. And they were like, ah, you know, you shouldn't care about that. Like, let us handle it.

Henrik Göthberg 17:01

You're an SM guy, you know, keep keeping your corner. That's when the whole idea is. A little bit, yeah.

SPEAKER_07 17:08

Out of frustration, but I mean, I guess not only as well, like we had this idea at one point in Campania to try to build this like business cognition type of company where like we optimize the business metrics end to end with like advertising as the lever. Um you remember this.

How Google Ads Auctions Work

SPEAKER_00 17:26

Uh we built our own tracking, and yeah, so we were kind of going in.

Anders Arpteg 17:30

And perhaps perhaps we should explain a bit how that type of SCM marketing really works, uh, campagna style, so to speak, and then if we expand it a bit more to what we do today. But if you were to explain a bit more for yeah, the type of second-order price marketing and these kind of techniques, how does that work really?

SPEAKER_07 17:49

Yeah, so you want to go? Should I go?

SPEAKER_00 17:51

Uh yeah, go ahead then.

SPEAKER_07 17:53

Yeah, okay. So basically, like Google has a site, a lot of people come there and they Google things, and like they started out without the business model, pretty much. They were just like, Oh, people uh like to find the structured information of the internet, yeah. But then at some point, like they added added this AdWords feature. Like, I think one of the first customers was like a lobster sales guy that was like, Oh, I can finally sell my lobster.

Anders Arpteg 18:16

Lobster, you think of something else these days, but yeah, lobster they keep coming back.

SPEAKER_07 18:20

It's them and the gremlins, I guess. Yeah, yeah. But uh, but then like they they took in like uh um I think he was like an economics professor, a halvarian, like microeconomics professor, and and he was like, You should really do this as a second price auction. And what that means is that you can basically say, you know, these clicks are worth this much to me, like 10 crowns or 100 crowns.

Anders Arpteg 18:45

And just to give some more background, I mean if you go to the normal Google page, you will see the organic results, like the real web pages, but you also see ads, and the way then then you place your ads there is by setting some kind of you build a campaign in in a Google ads interface.

SPEAKER_07 19:01

Yeah, you then add ad groups, and you then add your actual ads that will be displayed, the text, and also a set of keywords that will trigger. Like these days it's it's more complicated, but that's how it started pretty much. And these keywords match different uh sets of queries depending on settings, search terms, yeah. And and then basically you also set what a price that you're willing to pay for this particular keyword, a max price.

SPEAKER_00 19:27

Yeah, I mean, just like you would do on when when you do self-stuff on Tadera, basically.

SPEAKER_07 19:31

But but the genius of the second price auction is then that you can say honestly what is worth to you, but you pay the price of the next guy who hasn't the same value. So maybe the the most valuable click is 10 crowns, but you are willing to pay 100 because you earn a hundred per click, then you're still just gonna pay like 10.1 crowns per click. But if the other guy optimizes his business so he can pay 50, now you have to pay 50 to be on top.

SPEAKER_00 20:00

And on top of that, each landing page gets graded in a college score. So if you have a high quality score, you get cheaper clicks. If you have a low cord score, you will get screwed clicks.

Henrik Göthberg 20:09

Is it the same, but much more sophisticated today?

SPEAKER_07 20:12

Because this is the this started in the year of the keywords or search words, and now everything Yeah, it's fairly similar, I would say.ai. That Google was basically removing some of the levers that we had used really successfully when building campagna, and it felt like basically we were sitting in a very dependent position versus Google, yeah, and that window was kind of closing, so it felt like uh like we should probably figure out something else.

SPEAKER_00 20:51

Yeah, and I mean, yeah, Google is just removing levers to pool basically, and nowadays it's just basically oh, do an AI Max campaign, and then they do all the optimization for you. And of course, they're in the business of selling you bad clicks and good clicks, but Campania was in the business of only f finding the good clicks, yeah.

Anders Arpteg 21:08

Yeah, so yeah, they're automatic optimizing the way to set the bid. And I guess that's the hard thing with not knowing really what you will pay, so you could also try to gain that if you do it properly and get a higher position basically by by doing so.

SPEAKER_07 21:23

Yeah, and that was I guess the core innovation of Campania that basically we realized that if we just re-evaluate this very, very often, we can then pay the correct price at all times, and that's like high frequency building. And it turned out to be super profitable. I I think we beat Netflix's own engineering team with like 30% better results multiple times, even though they went back with our gonna build again and gonna win this time, they came back and we beat them again. So like uh it was a super successful approach, like very expensive to build. Like we had a big engineering org, we had like multiple people working pretty much full time building APIs, but uh also very Successful.

SPEAKER_00 22:00

And it was kind of hard to explain to customers as well. You had to kind of go into this PhD, Beijing, and math, and they were like, Oh, what the hell are you talking about? Yeah.

SPEAKER_07 22:08

But I think that that was like not in terms of technology, maybe, but it was your most valuable contribution for campaign as a business owners when you took this 3D chart of the value over time.

Henrik Göthberg 22:18

The Dov tot.

SPEAKER_07 22:19

Yeah.

SPEAKER_08 22:20

And then they finally got it.

Henrik Göthberg 22:21

Tell us this war story. This war story in a nutshell. I want to hear this.

SPEAKER_07 22:29

It was like this, then if I should do it shortly. So I had built this high-frequency bidding engine. But the problem was that we were very naive when it came to what the value of clicks was. We could control the costs, but we didn't know the value. So then we looked at the market. What is the best talent to help us solve this? We found Anders and AI PhD already back then, which is, I think it wasn't, they weren't growing in trees. Even when we found some more later, like not everybody was as good as you. So that was a fantastic find for us. And you helped us build basically AI models, what was called AI at the time. This is, of course, like an always changing definitions.

Henrik Göthberg 23:10

What was the underlying tech?

SPEAKER_07 23:12

Don't focus on that. But and out of that model also came a graph that visualized the value over time. And that is something that the customers finally understood that okay, if I had the same beat all the time, it's not following the change.

SPEAKER_00 23:24

And it's this beautiful, you know, it goes from like uh red to green, and it's almost like a landscape. So I could basically show Netflix this like, oh, when do you think your users are most likely to convert? And then you could see this kind of 3D landscape where it's Sunday at eight o'clock in the evening, the graph just jumped basically, and they understood it. And it was a I mean, a really innovative product, but it was also a very good way to explain. Visualizations and the way how we communicate. Exactly.

SPEAKER_08 23:54

Although it's such a big leap, if you can't explain the fantastic product that you built, like uh you're not gonna make much business anyway.

From Bidding Engine To Growth Engineering

SPEAKER_00 24:01

My deck went from like uh one hour to one slide, basically, and then it just had Netflix on the top, and then you're exaggerating, but yeah, good.

Anders Arpteg 24:10

But uh okay, you have expanded a bit. If you were to just perhaps explain the business model of growth hackers and cogniz compared to campagna, how would you differentiate it?

SPEAKER_00 24:20

Yes, I mean if if campagna was very deep in one channel, uh Google Ads, uh Growth Hackers was more taking another approach, which is you need to do everything fairly right in each step of the entire user journey. So from the channel into the site and all the way down to revenue, in order to get a good funnel and get kind of good growth going. So, what Growth Hackers did was to help orchestrate this to get like the uh CookLads campaigns or all the campaigns uh trackable and testable, and then get the user journey on the site, all that tractable and testable, and then be able to run experiments in all these different parts of the user journey and do that as much as possible. So the more experiments you run, the better growth you got, basically. And we call that the growth process.

Henrik Göthberg 25:08

And and we we talked about it before at that point in time, growth hacking, yeah, but this is now more or less established in around the people that know. So we we call this growth engineering now, or yeah, yeah, or go to market because it's a little bit different if you're an older it's part of GTM.

SPEAKER_00 25:27

It's popular right now, at least. Yeah, you know, marketers they go like it's like fashion.

Anders Arpteg 25:35

Passwords. Uh okay. So what's the difference between growth hackers and cognitive?

SPEAKER_07 25:40

No, but that was the fantastic thing. Like I we had tried out AI in different forms throughout our years. Like, I I I don't know if you remember when this like AI dungeon game was launched on top of GPT-3. I mean, that was the real first indication I think that that these new transformer models could deliver something special.

Henrik Göthberg 25:59

When is this? This is before Chat GPT, of course.

SPEAKER_07 26:02

Yeah, yeah, exactly. So that's like I think it's the summer of 23, something like that. Or maybe it was Christmas of 22. Christmas. Oh yeah, sorry, sorry. Yeah, so like is it Christmas of 21, something like that?

SPEAKER_00 26:15

So this is like the IPSIC now. No.

SPEAKER_07 26:19

Alright. Yeah, so okay, so they shipped that one. I actually applied for the API, tried it out a little bit, but like quality was too bad. But yeah, no, sorry, that was of course 2021.

SPEAKER_04 26:30

Yeah, yeah.

SPEAKER_07 26:31

And then as ChatGPT was launched in November 2022, then it was like, okay, now they managed to get it good enough, and in a pretty short time window. So then we basically just uh went all in on Cogni and I kind of we were in a pretty like hectic spot with Growth Hackers at the time, so we didn't have like time to work it on it on the day-to-day. But then as kind of Christmas break started, I I pretty much uh coded together an MVP and gave it to Tom for Christmas.

SPEAKER_00 27:01

I was very happy. I told all my whole family this fantastic Christmas present.

SPEAKER_07 27:05

What year was this a present? Yeah, 22.

SPEAKER_00 27:07

Oh, it was then, yeah.

SPEAKER_07 27:08

Christmas 22, yeah. And then like early 2023, we shipped our first like uh SQL to uh our like AI-powered SQL generator compilot thing. I mean it was pretty worthless because you more or less had to know SQL to be able to generate SQL, even though you write natural language, but it helped me at least. So we had a user of one basically. But then we kept building. Um, and the idea was to basically take uh what we do by hand and do it with AI.

SPEAKER_00 27:40

And I think it gradually kind of took over more and more of our work. I mean, in the end, we basically just got questions from clients. We fed it to Cogni, and Cogni created a nice analysis that we then presented.

Building Cogni From SQL Copilot

Anders Arpteg 27:53

And was it mainly to generate the SQL itself, or did you do something more with visualization and stuff? Or what's really the core functionality of Cogni?

SPEAKER_07 28:01

Yeah, so it's evolved a lot over time, of course, with capabilities. Like we we have tried to trust AI a lot, and then we have been like burnt. So, like the the first step was obviously like this AI co-pilot for your BigQuery data warehouse. Like, you ask the question, the AI delivers, and it's kind of still that in the core today, but it's obviously much more capable now. But then we tried to build like uh if you when deep research was uh released, we tried to build like a data analyst service based on the same principles for BigQuery, but like the analysis that came out weren't that good.

SPEAKER_00 28:38

It was hallucinating. I mean, if it's hallucinating a little bit when you're building a web page, maybe that's not you know it won't destroy the compl the entire page. But if you do an analysis that shows you know you made three million instead of 300,000, that's a little bit bigger, bigger problem.

SPEAKER_07 28:54

But but still, like this was in the same era, like we uh we have um the we know Anton from before, like so. We we met him. You mean Anton in Lovable? Yeah, exactly. Anton was a guy, yeah. Yeah, so so we met him around that time. He was hacking on GPT engineer. Uh yeah, that's right.

Henrik Göthberg 29:09

It's almost similar timeline.

SPEAKER_07 29:10

Yeah, and he was like, you should try it for your growth business, and we did like we generated some like uh dashboard pages and stuff, but we were like, nah, it's not good enough, you know. Fuck that. Yeah, so so it was like at the time where models weren't super capable, yeah. So then we kind of took a step back and instead built like uh more static reports where like the human could define the SQL queries, like make sure there was no hallucination in the core data, and then do like AI analysis on top of that. So, like, what does the data actually tell me?

SPEAKER_00 29:47

Um because I mean, if you present a dashboard to a client, the next question is always going to be but like, what does it mean? And that's basically what you can use an LLM for to give you some you know bite-size insights.

SPEAKER_07 30:01

So and then uh I don't know how how far we should go.

Henrik Göthberg 30:04

No, but we we talked a little bit upstairs as well, and I also a logical progression that okay, now we can put produce all this report, but no one fucking reads the report.

SPEAKER_00 30:13

So now you're now thinking of the next agent on top as the but I think let's I mean pause at that a little bit because that was like a huge surprise to me. I mean, I'm like uh think that everyone wants data, everyone wants insights. I don't care in which form it comes. But when we sat there and used cognizant to produce insights for us, and we went there to present it, everyone was like, Oh, fantastic! How would you find this? And then we were like, Okay, but you can get this every week for your specific role. Uh even everyone in your organization can get their own insights. You can go to the Monday morning meeting, and everyone knows how last week won, and you can get some recommendations. So we started sending emails, but nobody opened the emails, which is kind of interesting. Interesting, yeah. Because I mean, insights should be insights in my world, but they're not.

Anders Arpteg 31:00

So you went basically to do some vibe analytics in some sense. Yeah, exactly. We actually, yeah.

SPEAKER_07 31:06

And then like at the time that Lovable really started working, I think, and they had rebranded and everything, and that was I think summer pretty much of 24, right? So they they launched for the public autumn of 24, but already in the summer they had like something that started to work much better, and that was obviously Sonnet 3.5. Yeah, exactly, exactly. Like that that base capability then unlocked like a good enough user experience that could be used.

Anders Arpteg 31:35

We actually had Anton um on this podcast right before he launched uh Lovable. So he hinted a lot about what was.

Henrik Göthberg 31:40

Yeah, so we we we we take the claim. If you want to have a really good rocket ship, go to us and see what happens one month later. This was like one month before he did the product hunt launch for the brand. So it was so fun. So we take all credit for this. Congratulations. After Sana was here, they were able to sell it.

Anders Arpteg 32:03

Let's see the train. And he was much more open with how it actually works. Was Max here as well? Or no?

SPEAKER_00 32:08

Okay.

Anders Arpteg 32:09

That's an outline.

Henrik Göthberg 32:11

You're in for a good right now.

SPEAKER_00 32:12

Yeah, okay.

Henrik Göthberg 32:17

With um you were making a point here at this point in time, Sonic started to work better.

From Insights To Actionable Tickets

SPEAKER_07 32:22

Yeah, and that kind of unlocked like real analysis again, like that the copilot we had built before all of a sudden started kind of working. So then we went back, and and by the time we had kind of re-engineered the the analysis, even newer models were out. And we were then like going away from this human-built template to like a more of an agent uh spec, like you should do this and analyse this area. And and we were running that for some time, and then around I think Christmas, we felt that okay, this like agent-driven analysis. Like the agents get an area to analyse, but it doesn't get like a specific query to run or something like that. Those started getting good enough quality that you can actually take action pretty much straight from those. And at the same time, users weren't reading the reports anyway, even though like the quality got really, really good. So then we built this whole like uh ticket orchestration system with actual agent execution on top of the insights from the reports. What do you mean?

SPEAKER_00 33:26

So people I mean, basically, if you go back to the example, like let's assume that everyone in your organization has insights. What's the point of that? What are you gonna do with it? Exactly.

Henrik Göthberg 33:38

You're gonna you know create an experiment or act on it somehow. So you want a recommendation you should think about this type of A-B testing for your your role in SE. Like depending on what role it is, it will look a little bit different. What they should be doing.

SPEAKER_00 33:51

Yeah, so we then created this ticket view that we call it, or and and that's another agent that then uh basically reads the insights that the insight agent created and then create actionable tasks. Like we recommend you should do this, we score it with the eye scoring, where you basically take the most uh the highest impact uh ticket with the highest value and prioritize that first and present it in the ticket view that goes down from that's cool, yeah.

Anders Arpteg 34:16

And I think you mentioned also that I mean one approach would of course be to have one big agent or AI that does it all, so to speak, or you can split it up into different agents with different purposes. And I think you chose the latter, right?

SPEAKER_07 34:29

Or can you yeah, I describe more. Yeah, it's yeah, it's of course you could think of it in many ways, but like the tickets are really like uh a specification that uh a ticket execution agent could then take and run. So it's both kind of what we need to do, what we expect should be the outcome, and the why. And the why goes back to these uh reports that we had generated earlier. And it also has like how do we intend to do this. So it comes with basically an ask for permission to you know maybe publish a new page on a blog or maybe change budget on some campaigns, or maybe create some new campaigns and send traffic to the new blog post. And if you approve that, the agent goes ahead and executes.

SPEAKER_00 35:13

Because I mean we started with this uh ticket view, we're like, oh, this is cool, but then we started to assign ourselves, okay. And that's not so cool, yeah, exactly. Because I mean I'm not gonna have an AI telling me what to do, and then the AI basically tracked what I was supposed to do. So then we were like, okay, let's just have the AI execute this as well, and then the whole MCP uh trend came. So we started playing around with that.

Henrik Göthberg 35:37

So then so then of because now you get into thinking about recommendation, but then actuation. Now we're getting into actuation space, right?

Anders Arpteg 35:44

No, but but keep to the topic here. I think it's important you know, thinking here from a more architectural point of view. Should you have one agent to do it all, or should you have separate, more uh specific agents that have specific purposes? Or can can you go a bit more in depth?

SPEAKER_07 35:59

Yeah, like I I think we always have the context window to to think about, of course. Like it's not infinite, it's one million now, but like that's easy to burn. So if you if you specify like a sub-agent's task really well, sub-agent could execute it, maybe even with a weaker, cheaper, faster model, then then what you need to actually ideate it and like make sure that this makes sense to do. I mean, this is essentially how human orgs are kind of supposed to work as well as well, cognitive load, ultimately, yeah, as well. And also, of course, inside judgment context, like I mean, the CEO knows the direction that is going and could take like we need to do this, delegates it to somebody that has like an idea of kind of how to do it within their specific area, and in the end it it maybe lands in an expert's hands that only thinks about how to make it.

Henrik Göthberg 36:47

So, real intelligence in an enterprise is distributed, that's the point, right? Yeah, then no one can have uh can understand from direction to coding decision.

SPEAKER_07 36:56

Yeah, and like maybe AI could do that, but you also want parallel execution, you want like idem potent tasks that get like you want to impose judgments on different layers as example, right?

Henrik Göthberg 37:08

So, how do you if you have no way of going in and do we?

Anders Arpteg 37:11

I mean, that's a I think it's a question, right? It's uh it's not obvious, I think.

Agent Architecture And Harness Engineering

SPEAKER_00 37:15

But I think for us it was more like as we build out the product, we needed to do it in steps in order to understand that it was doing practical decisions. Yeah, exactly. I think so. And I think also today when we sit in demos and explain it to uh marketers or CEOs, I mean they explain they understand it directly because it mimics a human process basically.

Anders Arpteg 37:35

Can you go a bit into the tech stack or using like Langgraph or something?

SPEAKER_07 37:39

Or no, no, not nothing like that. We're like we build in TypeScript, Mono Repo, pretty much. Okay. And um, and yeah, no, then it's our own agent harness that we yeah, we haven't talked about uh that's that's the new fancy word, harness engineering.

Henrik Göthberg 37:56

You built your own harness. What are you what are you talking about?

SPEAKER_07 38:00

Yes, I think. But like what people talk about is of course that like the the underlying models themselves have like great capabilities. On top of that, uh obviously like Anthropic built Clode Code. Cloud code is a set of prompts, it's a set of tools, and they have the added benefit that like all of their RL is probably done with the same set of tools and prompts, so like it becomes really really strong as a companion to the model. And that that tool is is made to you know to fulfill coding tasks pretty much for maybe a single terminal session at a time. Now obviously they have web and all of that, but but that's roughly what it is, right? Uh a set of primitives for the alien to to use, like from the token generation, like now I should emit the tool call, it looks like this, I expect this to come back, etc.

Anders Arpteg 38:53

I mean, I I don't think people you know recognize sometimes the value or the USP that the harness or the scaffolding or the logic around the model actually do have. I mean, I think Sam Altens once said something like Um if your startup or your company doesn't get better as AI improves, then you're on the wrong track in some sense.

Henrik Göthberg 39:14

Yeah, but I really want Bernard to go deep on harness because I think this is one of the biggest misconceptions by people who don't understand AI engineering, that we haven't we have a we have one type of evaluation criteria for a very large broad language model, it should be general, right? And now we want to use that generality in a specific context, you know, for whatever we want to do, optimize growth or whatever. And all of a sudden now, if we are not putting a harness in relation to the e-val criteria we have right now, it will hallucinate by our new standards, right? Uh and and I think this is so so this is becoming super clear, I think, in Silicon Valley, or you know, we had uh we had the guy from Google here, Magnus, right? You know, who who's who was running the eval criteria for Gemini, right? Thinking about how the fact we put eval criteria on something that's that is general, yeah, impossible, you know. So I think this fundamental.

Anders Arpteg 40:10

I mean, one thing is to put cardrails on it, and another is to put the logic on it, right? Yeah, and what one do you actually do? Exactly. What is this?

SPEAKER_07 40:18

Yeah, I mean we are our own set of tools, obviously. Like we have a loop that goes around the call that goes to, for example, Anthropic that costs us money, right? That that's what the hardness is at the very core.

Anders Arpteg 40:31

Yeah, but I like you the loop is one thing then, and you have to have some kind of logic in you know when you should stop or when you should uh do some other calls and HP.

SPEAKER_07 40:39

Yeah, of course, like what prompts do we have, what tools do we have, what type of like uh uh descriptions comes with the tools, are skills auto loading or not? Like what happens when the task is done? All of this is kind of orchestrated into the hole that is like our cognitive cloud offering, where like as the task gets delivered, if it's a coding agent task, it becomes a pull request. If it's uh uh, for example, advertising task, it becomes like uh an analysis ticket in the end, where like another AI takes over, looks at what the first AI did and tries to do an analysis. If this analysis comes to the conclusion that it's good or bad, we log that. If it comes to the conclusion that this needs to be analyzed in some time, because like SEO takes time, campaign evaluation takes time because you want statistical confidence, etc. Like all of this is the harness.

Henrik Göthberg 41:33

But me me now, who really I'm really trying to understand and learn this because I'm not an engineer myself. So are are we distinguishing between guardrails, harness, or scaffolding? Are we using them interchangeably? Because I think you you were highlighting, right?

Anders Arpteg 41:47

That there is more there is It's easy to think that harness is about guardrails, but I think in this sense it's much more harness is more the scaffolding than I think about guardrails about it not fucking up, basically.

SPEAKER_00 41:58

Part of it, right? Yeah, but harnessing is way more. And I mean, yeah, yeah, I would say so. I mean, that's more about creating the value or distilling more value from a general LLM. Um but the the guard rates is like you're not allowed to change the case.

Henrik Göthberg 42:12

Guard rates is one thing, but a harness should be very tightly connected to the eval criteria for the objective function you're trying to solve, right? I mean, like so it's you so uh so when you say harness, it's like you should be doing this, but we also want to evaluate is it is it's becoming better? I don't know, is that part of the harness as well?

SPEAKER_07 42:29

I mean, we don't train our own model, so we don't have a like the objective function loop in our own RL. We could do that, but it would probably be kind of expensive, and the question is like, can we keep up with uh Frontier Labs? And like I think today the answer is probably no. Like at the same time, we're building quite interesting data set with like this our marketing experiments that have resulted in growth or not growth, for example.

Henrik Göthberg 42:53

Like, where do you put the memory? Stuff like this. All is that that's our harness, yeah. That's your harness, right?

SPEAKER_07 42:57

Of course, like we we build like an organizational memory for the customer's account. We try to figure out like what is driving growth for them, what is not, what should they do, what should they not do, like what's the positioning in the market they should like lean into or lean out of. Like all of this is orchestrated within the harness.

Henrik Göthberg 43:15

This is so much more. I mean, like, so I like I love the definition AI compound system that uh Berkeley AI Research Lab, I don't know, they put in the paper a couple of years back because it's it highlights the the there are different things that need to work in in orchestration here, and it maybe the harness way of thinking about this is one way to get more people to understand what this is all about. I don't know. What do you think?

Anders Arpteg 43:38

But one way to speak about this is also to use terms like um yeah, going from software engineering to agent engineering, meaning it's more about the engineering you're doing is really trying to orchestrate the agents properly, uh and then perhaps seeing how you can get them to collaborate properly to do what you want them to do in some way. Yeah, no, for sure. We're leaning into that like. Like madmen perhaps that could be a term to use potentially.

SPEAKER_07 44:06

And what we launched recently is something we call YOLO mode. YOLO mode.

SPEAKER_00 44:13

Yeah, because I mean this was another like uh feeling when you sit there and you kind of just approve ticket, approve ticket, approve ticket, approve ticket is like why do I need to sit there and just you know smash smash mash mash smash? Maybe it should just auto-approve. So then it kind of uh scary.

Anders Arpteg 44:28

Yeah. I saw this podcast recently of uh Andrew Caparthy, and he you know he doesn't like the term vibe coding, even though he really invented that or coined that term. He said, you know, in in the beginning, you know, since he's a programmer since very since very young age, he loved to actually program and he wants to see the code and he wants to review what the AI is doing. But it came a point very recently, like December last year or something. Exactly, yeah, where you know he said, you know, the AI does it right every time. There's no point for me to even look at the code. Yeah, so I become what I don't want to, a bytecoder.

SPEAKER_00 45:02

Yeah, I mean I had that exact same moment with Cogni actually, because I mean uh we worked with um you know 10 years uh training growth hackers and becoming these kind of uh innovative uh business people that know technology and marketing, and they are supposed to come up with growth hacks or ideas how to grow a business. And of course, I've uh been pitched a lot of these ideas uh through the years, and um just in December, actually, yeah, Cogni started by chance. Which underlying model?

SPEAKER_07 45:35

Uh 4-5, opus four.

Henrik Göthberg 45:39

Yeah, exactly.

SPEAKER_00 45:40

Yeah, so I we were sitting like looking at in the ticket uh board, and we saw all these experiments that uh Cogni suggested to us, and it was like, oh, this is good. Uh that's that's good. Where's the hallucination? Oh this is actually better than my growth hackers I used to train. So I mean that was our holy shit moment basically.

MCP Explained With Real Examples

Anders Arpteg 45:57

Yeah, that's so it's really cool. You mentioned, by the way, uh a bit about MCP as well, and I think for the people that are a bit tech interested here, it would be fun to see you know how are you making use of MCP uh today?

SPEAKER_07 46:09

Yeah, MCP is is kind of like USB-C, right? It's just the standard to connect something. And for for the longest time it it like did have some momentum, but now I think it's really gathering momentum, like at an insane level. Uh so we basically leaned into that. We build all our own connections to other platforms as MCPs, we federate those MCPs into our own MCP.

Henrik Göthberg 46:36

And what is MCP standard for? And what are we have out and do we have alternatives?

SPEAKER_07 46:41

Uh there are a few that are like semi-alternatives, I would say, like agent-to-agent protocol would be one, for example, proposed by Google, but uh like I would say pretty much no traction.

Henrik Göthberg 46:51

So MCP stands for model context protocol.

SPEAKER_07 46:54

Model context protocol is now owned by the Linux Foundation, but originally from Anthropic.

Henrik Göthberg 46:59

So and from Anthropic Linux Foundation, it becomes then the standard language uh connection interconnection. It's a standard protocol for connecting agent work.

SPEAKER_00 47:09

Yeah, the way that I explain it to business people is just that it's an API that an agent can talk to.

Henrik Göthberg 47:13

Yeah, it's an API and a protocol specifically for agents. Yes. Yeah.

SPEAKER_07 47:20

And on top of that, it has like authentication primitives, for example. So like I could come to Stripe, I could say I'm a Stripe user, and now I want to use your thing in my agent, then I connect to their MCP server, and I can then uh authenticate with Stripe using the MCP protocol. Like one thing that is notably uh missing is of course payments. Yeah, there is apparently like a pull request. Yeah, it's in the works, but it's not on the prioritized list of things that they will focus on this year.

Henrik Göthberg 47:51

But and if I understand this, like some oh, I have never done uh MCP engineering, but I know API engineering really well. I've been using APIs all my life. Is it different?

SPEAKER_07 48:02

It is because the MCP is kind of a little bit more alive, it's more agentic from an API point of view, if you will. So the a typical API, I would like to expect you to give me maybe your swagger spake or an open API definition or something like that. I'm gonna go off as a happy developer and hand code it. Like now we're in the 2010s era.

Henrik Göthberg 48:23

Yeah, yeah, I have your documentation, I have all this, and now I'm not sure.

SPEAKER_07 48:25

I might trust that you do exactly what you say, and then I can do exactly the same, and like we're all happy. Yeah, but model context protocol can like change endpoints depending on context, can provide example prompts. So they can be more dynamic with each other.

Anders Arpteg 48:39

Yeah, yeah, they can't, I guess, also less less demand on humans, so to speak. In an API, you normally have to have some kind of coding being done to use the API.

SPEAKER_07 48:51

Yeah, this is just plug and play, like the agent figures out in real time how to use it.

Anders Arpteg 48:55

They can just take a URL to an MCP and add it to another agent, and it starts to make use of it without you having to do any specific coding to use the code. Yeah, exactly.

SPEAKER_07 49:03

And now there's even MCP apps. So like the MCP URL can then like inject UI that optionally could be rendered by the MCP client. So that's really cool. Like, I mean that means that for example, in Cogni we have this concept of core metrics. Like we we ask the user to obviously, with the help of AI, define what they really care about, and then we're gonna optimize that, of course. Uh, but it's much easier if we know what they measure, and then this can become like your core KPI card, and we can actually design a little UI component. So if you connect your cloud cowork to the Cogni MCP, and then you ask, like, ah, how is my KPIs going? We can even send like a user-friendly version of that rather than just like the JSON numbers or whatever.

Henrik Göthberg 49:50

So when you start thinking layers on layers on layer, imagine what it what you could use for it, it's so much.

SPEAKER_00 49:57

Yeah, yeah. I mean, if you think about like I think the most people are in this kind of copy and paste mode with their AI, they're like going into this system, copy pasting that, and then pasting it into the AI sitting in the like back and forth and back and forth. And I mean the MCP is just completely destroys that, and you don't have to do anything anymore. You can just you know ask the AI to go and fetch your data in that system, crunch it the way you want it, and then send it off to that system.

SPEAKER_07 50:20

So and like a concrete example, like so yesterday there was this Y Combinator event. Obviously, like that's kind of our ICP. A lot of startup founders who want growth. We provide growth engineering as an AI. So then I asked Cogni to launch an X campaign, X ads campaign, targeting Stockholm, targeting people, uh like Y Combinator profiles followers. And it then went to um basically Nano Banana using MCP, generated a couple of ads, like we were ideating about like what vibe we should go with, etc. And then it launched a campaign on X.

Henrik Göthberg 50:57

And it was like, you know, I was cooking food for the kids in the meantime, so it's we are getting to a very different world now with people who are getting it and using these tools versus the rest. This is the the the gap is widening right now.

SPEAKER_00 51:10

Yeah, fast. And I mean we can notice that when we sit in in demo meetings. I mean we can like in five minutes in, we know what stage they're in, basically.

Henrik Göthberg 51:19

Yeah, and and how how do you how do you differentiate your selling or your communication with them when you realize someone is less mature and someone gets it? Because this is now getting this is Mars and Venus stuff.

SPEAKER_00 51:30

Yeah, exactly. I think it has to do a lot with trust, uh, because as this December moment, that's when you start to trust the AI basically for coding. And I think every human needs to go through that phase basically. So if we notice that they're not leaning in that much to AI, then we focus more on the kind of analytical AI part than the oh, you probably run a couple of campaigns here and there, and it's probably a hassle for you to you know, again, paste and copy all the data and create a report. So that can help you out. Yeah, exactly. And it's like you can start with reporting stuff, uh, your team will get more insights, and then when you're you know happy with that, you know, we can connect a couple of MCPs and then you can start executing basically.

Henrik Göthberg 52:12

And how does it sound when someone is switched on? Like someone is on it. They just want to know more and more and more.

SPEAKER_00 52:18

Like uh they just connect every MCP that they have and like let's go.

SPEAKER_07 52:22

We show the reports and they're like, okay, so this you I can really use this to execute yeah, wait, let us come to the next point in the presentation.

Anders Arpteg 52:29

And they hook up their GitHub account and yeah. What should we call this? I mean, if we have agentic development or agentic coding, is this basically agentic marketing then? Or what would be a good term?

SPEAKER_00 52:40

They cognitive coding as well. So I mean, uh yeah, I would call it like data-driven marketing and coding.

Henrik Göthberg 52:47

Same thing is it's coding as well, right? Because as you are coming up with an execution, you build just in time what you need to build. Yeah, yeah, right.

SPEAKER_07 52:56

Like obviously subject to human approval still. Like we don't have an auto-PR merger, but yet, yeah. Yeah, of course, yeah. But like, I mean, if for sure, in many cases, like uh so the solution could be partially changing the code and also partially uh doing actuation in different marketing systems or sending a new use letter, or like but because this is also a little bit I asked you before, are you for product teams or are you for marketing teams, or you're right on the intersect?

Henrik Göthberg 53:24

I think you I mean like you're right on the intersect, right? Yeah, yeah.

SPEAKER_07 53:27

Yeah, and I think like one of the core ICPs is really CEOs. Like so somebody said that, yeah, like I just hired a AI transformation guy, blah blah, and then I think it was Gary Tan on why combinator was like the CEO should really be the AI transformation guy. Like and I think it's true, like it's uh in the end, like it all boils down to that. Like it's it's where the the main agency of the company comes from or should come from, and you so I understand.

Henrik Göthberg 53:55

So when you say this is an E CEO place, like you really need to decide where you want to put autonomy agency and where you want to put people, higher guys, and where you want to put the in agents now.

SPEAKER_00 54:06

Yeah, but it's also like that's what you mean, right? No, but it's also that cognitive does all of these roles for you, so now you don't necessarily need specialists, you can just you know that's it becomes yeah, it becomes it becomes the it becomes the step to the one-man unicorn, yeah, and that's why it's a CEO question.

Henrik Göthberg 54:23

Yeah.

SPEAKER_07 54:25

And it could be like, I mean, should the marketing department decide your pricing strategy? But maybe your pricing strategy decides your marketing success. And this kind of converges usually on the CEO or that sort of thing.

Henrik Göthberg 54:38

Yeah, exactly.

Anders Arpteg 54:39

The interface.

SPEAKER_07 54:40

Yeah.

Anders Arpteg 54:42

I'm thinking if we should move into, I mean, you're both experts in uh search engine optimization as well and search engine marketing, but I guess nowadays, you know, you have um AI agents uh answering questions more than search engines. Yeah. Do you have any thoughts about this? Is this changing how you do the work? Uh if you want to get noticed these days in terms of sure definitely.

SPEAKER_08 55:04

I mean, it's that's all I think about. You know, how can I intern myself in the right context?

Henrik Göthberg 55:09

Yeah, how can we be found in the agent space?

SPEAKER_07 55:12

Yeah, yeah, exactly. And like uh who will use this data that I enter here, how?

Henrik Göthberg 55:17

Yeah, exactly. So what what's the answer? What's the fucking answer? No, but I mean it's yeah, yeah, of course.

SPEAKER_00 55:24

I mean you can go uh all different directions with this, but I think where we are right now in the market, you kind of just have to make your content readable for LLMs and structure it in a way so that LLMs.

Henrik Göthberg 55:34

So let's break that down. Step one, getting out of the search engine into the agentic mode, so ChatGTPTO, anyone can find you, you need to be readable, you need to be structured, you need to be code coded in a certain way. Could you just elaborate on that?

SPEAKER_00 55:47

Yeah, I mean it it I mean they like bullets, they like tables, they like things like that.

SPEAKER_07 55:52

So it's it's not they also like all of that to be server rendered.

Henrik Göthberg 55:56

So there's a couple of Tyson I I've learned about this. Like this, like server-rendered. Sorry, I'm I'm getting excited.

SPEAKER_07 56:04

It's exciting.

Henrik Göthberg 56:08

Sorry. This is fun. Yeah, sorry.

SPEAKER_07 56:11

Yeah, no, but I mean I the whole internet is being rebuilt as we speak, right? And and at some point, like somebody thought it was a good idea that the client computer should be big and should do a lot of computation. That's why my Chrome takes 60 gigabytes of RAM now, right? And that means uh the browser is like a fully built, super complex application where you can find like a lot of uh security holes, etc. And it also means that to consume this content, you need to execute JavaScript, you need to have a lot of ROM as well. But most agents are not full Chrome browsers, it's not a very like computer program friendly interface. That's why it's better to have like server-side rendered so that you take back this compute that you have offloaded to your customers and you render the page from day one, right? And they can just download it as text and parse out the information they need. On top of that, there are ways to structure the information uh with, for example, uh structured uh data standards like this LDJson format that makes it easier for agents to consume the data on the page.

Henrik Göthberg 57:17

So we're getting back to how you structure the web page and back to good old HTML rather than all these uh scripts and everything else.

SPEAKER_07 57:24

Yeah, it could be that, it could be like markdown versions being served. That's another track that people are taking. Like uh obviously with an HTTP request, you could specify which format you prefer to get the response in, and like you have the option to then, if they say they want it in markdown, actually give them markdown rather than all of these extra worthless HTML tokens that will make it look pretty for a human but will make it like look messy for an agent.

SPEAKER_00 57:49

But I mean these tactics aren't bad for SEO either. So I mean you you you could do both at the same time because LLM Googles as well.

Anders Arpteg 57:57

So I mean it's not bad to do SEO and you say Google, but you mean search of course, right? And and if you do use an um yeah, chat but chat GPT or whatnot, I mean they do still use the normal kind of search engines underneath, even though you you don't always see it. Yeah, but then is there I mean that there are similarities then as you say, that's so it's probably good to to optimize for both. It wouldn't hurt you to do that, but I'm just I don't know the answer, but but is there a way you know besides the format that you put on the page, how should you be recognized for when an agent do search?

SPEAKER_07 58:32

When an agent do web search, yeah, they Google in a very different way or they search in a very different way, they use more words, they they formulate themselves differently. Like they usually do this query find out, is what they call it. So, like you, the user, have one question, but they will do like 10 to 100 different web searches and visit a bunch of pages.

Henrik Göthberg 58:52

Different, a little slightly different uh prompts or questions they have.

SPEAKER_07 58:55

Yeah, exactly. Small aspects of this, what you're trying, you're trying to find the best uh apple tree to plant this time or you can do that.

Henrik Göthberg 59:02

They do 100 versions of that question, literally.

SPEAKER_07 59:05

Slight variations, and if you can start figuring out what these variations are, like commonly, you can then start optimizing your content for that. And there's a quite interesting metric, like if we want to talk hard numbers here, because most of the search providers aren't sharing data around this, but uh Microsoft uh Bing is doing that, they have a tool called Bing Webmaster Tools. You prove that you own the web page, and then basically you will get uh impression statistics, click statistics from their traditional search, and then they also launched an AI citation statistics page. Yeah, yeah.

SPEAKER_00 59:42

So that's the only official because all the other providers you see out there they basically come up with probable prompts, and then they run that prompt through all the LLMs and then they read the result. Yeah, and then they also pollute their own uh measurement.

SPEAKER_08 59:57

Oh, you appear more and more in the searches related to what you're monitoring how this is so interesting.

Henrik Göthberg 1:00:04

How many uh how do you look at your the industry that you're coming from with the CEO people? Because a lot of the marketing people and CEO people were not really so deep coders. Are they are they keeping track? Can they can they follow into this? Will we learn this, or is this a different type of uh agency you need to help get help from?

AI News Payments And Geopolitics

SPEAKER_00 1:00:22

Yeah, I mean I I think uh that has historically been what why you go to growth hackers, because you need all that technology to help you, and it's not necessarily you you're gonna see that on billboards or anything like that, but you get solid data you can trust, and you get like an infrastructure that you can run experiments in and things like that. And that's actually the the problem we're trying to solve with Cogni. Yeah, exactly. We remove all that scaffolding basically that you need you you need to have GA4 accounts, you need to have GTM set up, you need to have a good structure of your uh home page, you need to have you know all this. Now you can just uh get recommendations from Cogni, and Cogni will take care of all that scaffolding for you basically. Cool.

Anders Arpteg 1:01:03

And um it's time for AI News brought to you by AI AW Podcast. So we usually take a small break in the middle of the podcast and um just talk about the recent most exciting AI news that we can think of. Could we perhaps ask uh Tom and Bernard, do you have any news stories that you read recently that you'd like to share something about?

SPEAKER_07 1:01:32

I mean, what I what I read this morning when I woke up, it's always something, right? Yeah, it's always something. But but it was that Stripe had launched uh link CLI.

SPEAKER_05 1:01:42

Right.

SPEAKER_07 1:01:43

And and obviously, like CLIs are this like third leg of uh MCP alternatives, not A to A, but like the other way to do it is just that you ship a CLI that can do everything that you want the API to do.

Anders Arpteg 1:01:54

You're very technical, Bern, but I think you know CLI it's a terminal tool basically.

SPEAKER_07 1:02:11

And and why is link CLI then so cool? Yeah, it basically is a command line tool where we now can create payments that the user then can approve. And link is like Stripe's uh pay with one-click solution. So basically now I can give my agent the possibility to spend money.

Anders Arpteg 1:02:32

But this is sounds super dangerous, right? If you give open claw then access to Link C ELI, then they can make payments for you, right?

SPEAKER_07 1:02:40

Yeah, I think there's still a human approval element, but the whole like commercial layer of it is unlocked.

Henrik Göthberg 1:02:46

But the commercial infrastructure is there now.

SPEAKER_07 1:02:48

Yeah, maybe. So I think that now I can like programmatically tell an agent that like it's a payment you need to make, you need to input this into your CLI, and then it will prompt a human to approve it. So like I think the example from the CEO of Stripe was that like uh I asked uh you know if it was open claw to buy something for yourself, and then he had like a screenshot of the payment approval prompt that he got. And I think that like this whole agency commerce area is of course super interesting, uh both from like just a shopping perspective with like B2C products, but also like from a SAS perspective, like how can we enable agents to buy our services?

SPEAKER_00 1:03:32

But it it it fundamentally we we talked about it, the behavior around commerce is gonna get to into into a new arena eventually different world, and especially in marketing, because we are usually have to you know buy advertising spaces here and there, and if you can get that agentic, then that would open up a lot of media space for a lot smaller.

Anders Arpteg 1:03:56

If we we take Peter, you know, who created OpenClaw, he actually prefers CLIs to MCP, by the way.

SPEAKER_07 1:04:02

Yeah, he did add MCP now though. So he's uh we partially reverse that take, I think. But yeah, I know it's it's very common. Like we we think about building a cognitive CLI as well, it's uh actually already being built.

Henrik Göthberg 1:04:13

So what's the difference? I don't understand. Yeah, MCP-based or CLI-based? What is the distinction?

SPEAKER_07 1:04:21

So MCP is basically an HTTP protocol at the base. Like it's HTTP RCP remote procedure calls. Like it's a very definad way to build an HTTP API that then the agents can follow. But a CLI on the other hand, it could be anything, it's just a command that you can run and then it will print text and hopefully do something as well.

Henrik Göthberg 1:04:47

So it's even more open, more flexible in that sense.

SPEAKER_07 1:04:50

But at the same time, you need like a command line uh interface or like your own computer to run it, basically, install the command, etc.

Anders Arpteg 1:04:58

Open claw it's easier because it's it has a terminal, it can easily just execute the CLI and yeah, it's a bit more uh flexible, I guess, in that sense.

SPEAKER_07 1:05:07

Yeah, and a lot of services don't have MCPs. Um, and you can like if you have a coding agent, you can produce a CLI maybe more easily than an MCP. So that's why it's very good for open claw that it can basically tell the agent.

Henrik Göthberg 1:05:22

And your your line of thinking is to have both.

SPEAKER_07 1:05:24

Yeah, so we're gonna build a CLI that then calls the MCP. But it's serving both uh niches.

Henrik Göthberg 1:05:32

Yeah, yeah. That's a good story. Good one. Tom, do you have anything specific?

SPEAKER_00 1:05:37

Well, I do share this kind of geopolitical interest that you have on the show. I mean, I like the story about um the meta monus deal and how China just basically reversed that. Yeah, which I think is kind of hard to do because those people at monos are now working uh in meta. Uh and some more backgrounds. Yes.

Henrik Göthberg 1:06:00

Monus, what is what is the background here?

SPEAKER_00 1:06:02

Yeah, so I mean Monus is a uh Chinese model, or they started in China, they moved to Hong Kong and then to Singapore to kind of wash their Chinese models.

Anders Arpteg 1:06:11

Like a pre-open claw version, yeah.

SPEAKER_07 1:06:13

It's more of a harness than a model, I would say. You could plug in the multiple models into Manus, I think. But it's it's for sure. Like it, I guess it means hands. So it's like a closed hands, it's a similar idea, but more cloud-hosted than you run it on your own computer.

SPEAKER_00 1:06:27

And what was the fundamental pushback? Because China wants more AI power, basically. I guess that uh they I mean Meta, of course, desperately needed some type of AI hit. He Zuckerberg has tried a lot, and he's just buying stuff. Exactly. He wanted to buy it and got pushed back that nothing. He bought it, it went through, they signed everything. You know, the people went in, they even integrated monas into the meta advertising platform. And apparently, you know, the the revenue for meta jumped like 30% for a quarter when they did that, because the meta interface for us marketers is extremely complicated. Uh, so you having a conversational agent to talk to is of course fantastic. And then China kind of went in and said, Hey, this can't be happening.

Anders Arpteg 1:07:13

I mean, it's interesting how more and more of the trade or the collaboration is being blocked, you know, over the geopolitics here. Normally it's US that block access for China to get access to Nvidia chips and and uh other technologies, like the TikTok. Yes, the TikTok now it's reversed. Yeah, yeah. Now China is blocking US instead.

SPEAKER_00 1:07:33

So um, yeah, so it's it's getting, I guess, more and more polar polarized in the yeah, and back to what we talked about before, like this is being such a big change for humanity that it's not something you can really just leave. They have to pick the fight, basically.

Henrik Göthberg 1:07:49

Yeah, I think there's a there's there's another interesting twist here in this like who is leading the AI race story, who's blocking who. Now it seems like this is one other indication from that angle that they're doing great stuff in China, they're doing great stuff in the US, so it's not like one is always top dog blocking the other ones. No, it's much more similar.

SPEAKER_07 1:08:12

Yeah, no, and if you factor in robotics, of course, that's where they really are.

SPEAKER_00 1:08:16

Now it gets interesting.

SPEAKER_07 1:08:17

Yeah, conversation starts, I think.

SPEAKER_00 1:08:19

And of course, power. I mean, China has a lot more power than the US, so I mean electric power, yeah.

Cursor Deal Rumors And Claude Limits

Anders Arpteg 1:08:27

Yeah, okay. Uh anyway, uh super interesting future for sure. Yeah, and thanks for that one. Geopolitics plays out in coming years. That will certainly be interesting. What about you, Andes? Well, uh, this is not geopolitics, but it's more US politics, perhaps. Uh, but I think the acquisition or offer to acquire Cursor, which SpaceX did, was interesting. Yeah. So you can, you know, SpaceX, they recently acquired XAI, it's part of one, yeah, a set of these Elon companies, and uh and XAI had Grok as you know one frontier model. It's not been going super well for Grok, and it's I guess not having had the progress uh they were hoping. And he fired a lot of people from XAI recently, and uh it could be because of the acquisition in SpaceX.

SPEAKER_08 1:09:13

I think he said that it needs to be rebuilt from the ground up or something like that. But it's a strong indicator that something is not right.

Henrik Göthberg 1:09:20

So so uh one of Elon's companies is buying one of his other companies.

Anders Arpteg 1:09:26

No, no, but that was before. So SpaceX already acquired AI a month ago or something. Yeah. Uh but now's in AI time. But now then a week ago, then SpaceX put it off for up for Cursor. And Cursor is um you know very um uh famous coding environment and very popular in the enterprise business, similar to Claude and Cloud Code.

SPEAKER_07 1:09:52

Um, but I mean they they are the ones that define this tab tab paradigm pretty much like I guess it was in GitHub Copilot before, but they made it like wildly popular and has grown up to a bigger size than lovable and like yeah, yeah. But I think we've had this coding users on the show.

Henrik Göthberg 1:10:09

What's your favorite combo? Has been the question a couple of times. Oh, it's the cursor, yeah, yeah, no, cursor, anthropic. That that is the answer for many people as the main combo, right?

Anders Arpteg 1:10:20

And you can think about why he's doing this. One, of course, could be that Groc is not progressing at the scale that he wants to, and cursor actually are building their own models as well, composer, and and uh then uh some are rumoring that Anthropic is you know fighting Elon and they cut Elon or Tesla and uh SpaceX off from Claude, and potentially they have to have their own coding agent, and then Grok is not you know progressing in coding as as I hoped. So this could be a way for them to simply accelerate a lot, yeah.

SPEAKER_07 1:10:51

And it's also, of course, a treasure trove of data around coding, successful coding outcomes, etc. That is like you can't build that by himself because nobody uses Grok for coding purposes. Yeah, you could try to simulate it, but it won't be the same as like uh coding in the wild.

Henrik Göthberg 1:11:08

This this makes sense, he knows the benefit of data in his cars, you know, driving data in his cars. So now I want driving data in the coding.

Anders Arpteg 1:11:17

I mean, getting access to all the enterprise users that they have, but also the data they have provided throughout the years, of course, will help them to build. And then on the other side, XAI have the big Colossus data center, which is the biggest one in the world, more or less. Um, so then Cursor can use that to continue to build their models even faster, potentially.

Henrik Göthberg 1:11:35

But Cursor was separate, right? Who was the founder of Cursor? Wasn't there even a Swedish guy in there?

Anders Arpteg 1:11:39

Yeah, one of the four was Swedish, but he actually left Cursor now, so he's not part of it anymore.

Henrik Göthberg 1:11:44

But he was But have they been able to stay independent the whole way through until now, Cursor? Or have who who is who may who is he buying Cursor from? I don't know this.

Anders Arpteg 1:11:55

Yeah, I mean they they had I think three or four founding rounds or something, but they are still DC founded. Yeah, yeah. But I think the the interesting part here is also that SpaceX, you know, is rumored, of course, to do the IPO very soon for a the small amount of like two trillion dollars or something, which is insane.

SPEAKER_00 1:12:10

I mean it it's it's the IPO wars now with OpenAI and Tropic and SpaceX. And then they're saying that you don't want to be number two or three because the one that goes first is basically gonna take all the money that's exist in the world.

Anders Arpteg 1:12:26

So it's it it's insane how big it can be. And then given that it's going to happen rather shortly, and I'm sure they'll be doing all the filing preparation now for the IPO, then it's super hard for them to actually do this acquisition. So they're actually not acquiring cursor right now, but they are giving an offer. I think they actually got the right to buy cursor, but they're not going to do it now.

SPEAKER_07 1:12:49

Yeah, exactly. And I mean I think he is kind of Elon uniquely positioned to get value from buying it. So, like for most companies, it would be pretty expensive to buy cursor today for 60 billion, and like they are under massive threat both from obviously anthropic and claude code and then like uh more recently like OpenAI, GPT.

Henrik Göthberg 1:13:10

So he's the logical buyer, yeah.

SPEAKER_07 1:13:12

But like he is uh has the scale on the data center side or is planning to build it, so like for him it's super valuable.

Henrik Göthberg 1:13:20

But does it you said something um on the IPO? I just wanted to make sure because I thought that was pretty cool. You mean the one is not announcing it, but the the one executing it, you want to be first executing on this supermassive IPO. Is that what you meant?

SPEAKER_00 1:13:35

Yeah, because then the money dries up. Yeah, basically, I mean I listened to these financial analysts, and they're like, you know, so many people want to go into these types of IPOs. Uh, and if you do them, like it's like a uh movie premiere. If you have Avatar going up with another big movie at the same weekend, then people can't go to two movies at the same time.

SPEAKER_07 1:13:56

But it's also even worse because of obviously all of this cash is not sitting as cash, it's sitting as assets in other companies, and like they're gonna have to pull it from somewhere, and they can't pull infinite to put everywhere.

Anders Arpteg 1:14:08

So but imagine the capital they will have now from this IPO and then go to build terrafab, as we spoke about in the beginning.

SPEAKER_00 1:14:16

Um yeah, I think that's the way the reason why he's doing it. I mean, uh Eaddon has never had problem raising cash, but it takes time, he has to go around, meet uh investors, and and now he can just basically use the exchange for that. So I think it's with his plan, he needs this hose of cash.

Anders Arpteg 1:14:34

Yeah, and I think potentially the IPO could go up because of this, because it will create some kind of because Croc is not progressing perhaps at the same pace he's hoping for, and but with this acquisition locked in, so to speak, the price in the IPO could potentially go up. I don't know.

SPEAKER_00 1:14:49

Yeah, and I think also like 90% of the value of SpaceX is actually Starlink. Right. So I mean that's what people are investing in, and I think Starlink is a hugely fantastic business model. So yeah, yeah.

Anders Arpteg 1:15:04

Awesome. Good news. I think Goran doesn't have to. Any Goran?

SPEAKER_00 1:15:11

I don't have a mic. But shouldn't we talk about the nerfing of Claude? Uh no, please tell tell me.

SPEAKER_07 1:15:18

Yeah, so I guess the the vibe or the the rumor or what people are talking about is of course that this new like Claude Opus 4.7 is way worse than 4.6.

Anders Arpteg 1:15:29

Way worse.

SPEAKER_07 1:15:30

Yeah, it's it's like they and you if you look at like the the benchmarks that they published, like one of the things where they did uh some progress was in safety. And and this is of course related to the old Mythos thing as well, I think, that they started to think more about like the potential cybersecurity offensive uh capabilities of the models. But uh when when they've done like maybe that safety improvement, or if it's just like running it more quantitized on like uh yeah, cheaper, faster way to run it, basically, you decrease the number of bits for each parameter, then it has resulted in the model not being as good as it was a month ago. So I think the vibe, like you see a lot of people saying they're moving to codecs and GPT 5.5.

SPEAKER_00 1:16:16

I mean, the world on the street, if you if you look at uh data center investment, they're last. They have this deal with AWS, but that's gonna take and on growth, they were top.

SPEAKER_07 1:16:26

So this is the problem. Like they got them too successful for their own compute. Oh, that's hard. Yeah, and and I I mean I I think that there was this really interesting interview with Dario like a few months back where he was describing this dilemma where like if you overestimate your amount of customers, you have to basically invest so much that you're gonna go bankrupt. And if you underestimate, like um then you won't have any compute to train the next model, and now they're trying to thread that needle, so like giving less to inference, they still need to keep training the next model, but users are noticing because there's just so many of them now.

Anders Arpteg 1:17:07

It's a fight for the compute, it seems like uh was it's the um it was who was it? Uh Jensen Huang, I think, in Nvidia CEO said the five layer cake. Uh it was the um energy ships, uh compute infrastructure model and application, I think, right? And um, if you lack or have a bottleneck in any of them, you will be, you know, screw or yeah, screwed, more or less. It seems a lot are bottlenecked on the compute layer uh right now.

SPEAKER_07 1:17:37

Yeah, and will continue to be, I think. I mean, Elon Musk has said that he thinks that in 46 months the cheapest place to build computers to will be space. Yes, and I mean it's space data centers. Yeah, a lot of a lot of like uh uh linear curves on log log plots uh point to that, right?

Anders Arpteg 1:17:54

But also energy then, but then being in space, you basically have infinite inner energy from the sun as well.

SPEAKER_00 1:17:59

So and you don't have to cool it. Yeah.

Anders Arpteg 1:18:02

Uh well yes, I think you do, by the way. Do you do you have space? No, it's it's super hard to cool, I think, in space.

SPEAKER_00 1:18:08

Is it? Yeah, so you know space it's cool.

Anders Arpteg 1:18:10

No, I you can't transfer the heat out. It's it's kind of tough actually. Okay. But I think he has some interesting techniques from all the Starlink satellites you already have, so he knows how to do it. Yeah, exactly. But it's not that easy, apparently.

SPEAKER_00 1:18:21

Okay, but I I heard somewhere that he needs to um also send like a rocket every five minutes to get them out, which is insane.

Patrick Couch 1:18:30

Yeah, totally.

Henrik Göthberg 1:18:32

But sometimes now we just wonder: okay, IPO this uh terrafab that is it just bullshit, or is it what is it any sense in this, or is it just blown out of proportion and we're gonna see something different in reality?

SPEAKER_07 1:18:46

But what do you think is driving him? If not like to me, it looks like his decisions are mainly based in this like almost existential dread that like we're gonna same fuck it up.

Henrik Göthberg 1:18:58

Yeah, I think maybe he as a person is driven by those kinds of motivations more than normal people.

SPEAKER_07 1:19:06

He's taking on the mission to make life multiplanetär. Like, if his main motivation was to be able to drink like pinna caladas över day, then he could just stopped the palpit.

Henrik Göthberg 1:19:15

But okay, so let me reframe the question. I have no doubt he is actually completely sincere in the intention and what he wants to do. But is there a is there a feasibility in it, or is it something you can aim for and have that as your hardcore goal, but shoot for the stars, but but and and you will reach the treetops. What's the reality here, or is this the reality? That's my question more. So I'm not I'm not questioning his in being insincere, I'm just questioning where will it land.

SPEAKER_00 1:19:47

I mean, these tech CEOs are probably, I don't know, not the best people on the planet. But but I they do see farther into the future than we do. Yeah, I definitely think that. And uh I think that uh the whole Claudner thing is a tell that people are using AI a lot. I think that I I read another analysis that the like the revenue for coming from uh anthropic they can't just be coming from like IT budgets, they're like eating into HR budgets now. Um if you you know you can't grow that fast, basically. So I think we're it's moving faster than we think. And I you know, we all live through the IT eraa or the internet era, and everyone was like, oh, it's gonna find cars and everything, and then you know that didn't really happen. So maybe it's kind of that situation, or they see something that we don't.

Anders Arpteg 1:20:34

I mean it's a good old saying, like you always underestimate the short-term impact, but uh underestimate the long-term. Yeah, maybe it's that simple, right?

Henrik Göthberg 1:20:43

I was thinking exactly like this. Like we were overestimating how fast the internet would grow, but in reality, 15-20 years later, everything and more happened, right? Exactly. Except if you're Ray Kurzweil, and then it's like just on timetable.

SPEAKER_01 1:20:58

Yeah, exactly. He's nailing it, by the way. Yeah. Cool. 29, right? What 29. 29, yeah, I suppose AGI.

Anders Arpteg 1:21:10

I think you mentioned Banner and Tom, something about you know, code bases may be replaced by CMS platforms or something in uh in the future. Can you elaborate what you meant by that?

SPEAKER_07 1:21:21

Yeah, I I think that's the other way around. So basically, like CMS content management systems was basically built as a way to enable non-coders to build out content on websites to manage websites, and it helped humans back then, but now it's actually in the way because they add so much extra scaffolding and limitations, where like basically uh AI natively speaks code, like maybe even more than language, even.

SPEAKER_00 1:21:50

And if you take like the the SEO and geo perspective on that as well, when uh Google is supposed to read a site and it's a WordPress site, it has a lot of other code that's not maybe relevant for the specific site. If you go like pure code, then there is no code uh kind of unnecessary there.

Henrik Göthberg 1:22:08

So it becomes faster, it becomes just easier to read, you rank better, it's just basically so what you're saying in in layman's terms, we used uh CMS systems like WordPress or others in order to basically improve our speed of building websites. And when we would say democratize it, we democratize building websites to designers or whatever you want to call it. And now in reality, we can democratize in another way through through coding assistance and we build real websites in code that way instead, and democratize it even more because like all CMSs have like limits in terms of like what you can do. And what is then the ultimate if you now what if if you let's let's follow that trajectory. I love that idea. Where would you go as a designer who can't code now then? Would you continue with WordPress or would you go into Lovable, or how would you go about or use cursor and sonnets?

SPEAKER_07 1:23:02

I think cloud design is pretty good, like you could you could do that.

SPEAKER_00 1:23:05

Or you can continue working with Figma, which is you know, and and you can just have the MCP with cloud code, and you know so the logic is Figma, maybe, and then with Figma into how yeah, and use kind of cloud code, connect Figma, and then you can uh you know publish your sites. Because I mean when you ask uh Claude to generate a site, you get one of the their templates, yeah. Uh and and you kind of have to push it rather hard to do something unique. So that's what Figma is still strong at. Exactly, yeah.

SPEAKER_07 1:23:34

But that's at the moment because like Figma is still also kind of like a CMS for design, right? Yeah, it has a framework for like what you should put in, uh set schemas, and like they have built this whole product with the with some ideas in mind how it should work. And obviously, code is more expressive than that, and like with the the coding capabilities just scaling, like Claude or some other model will win with code in the end.

Henrik Göthberg 1:24:02

But this is vastly interesting how this plays out now because I I I've been onto this conversation even for my own company, right? Yeah, where we are not uh growing, um we are very careful of how and what type of uh competences we're growing and where we are building stuff right now. And the the core logic that we sort of installed was we want we want to follow the trajectory of the core coding tools. So, okay, so where is our content management right now? Oh, we are thinking about a GitHub, right? Yeah, yeah, yeah. You see what I mean, right? Of course. So we we are building our we and okay, everything we write, let's write it in Markdown immediately. So I'm I'm writing my daily blah blah blah in Markdown, and then I'm taking different things. And this is what I'm this is exactly the point. Should we go WordPress? Should we go HubSpot? Or no, no, no, we should go somewhere else. Yeah, very, very interesting for the a very big portion of the market.

SPEAKER_00 1:24:57

Yeah, I think I mean I read the statistics that uh WordPress is like 43% of all the sites globally.

Henrik Göthberg 1:25:02

Yeah, that's this is the statistic I'm thinking about, right?

SPEAKER_00 1:25:05

Yeah, but that also makes it makes it a target. I mean, we had a customer that got hacked, yeah, and their main uh page got uh basically redirected to a gambling site, which meant that they got uh thrown out of uh SEO and Google and you know, so being on a platform like that is also a risk. Whereas if you have your own code, you can do.

Henrik Göthberg 1:25:26

But what is your best? I mean, like you're gonna advise me. I'm gonna buy your consultancy for my for Dereducks. You know, how are we gonna think about this? So we're gonna go uh document, you know, the same way as we do code, we do text. No problem. MD, blah blah blah. And then we're gonna go. Uh no, HubSpot was the first one that went to. I want to go more open source. I want to go something which is fundamental open source coding frameworks and build my own code. I mean, I React, uh, you know, what am what am I doing?

SPEAKER_07 1:25:53

Yeah, I think next is a pretty strong candidate, which is of course like a React framework. Yeah, you could also do like Astro is another that has like excellent uh track record in terms of page speed, so you could do that.

Henrik Göthberg 1:26:04

But this is this is going the code route rather than the WordPress route. Yeah, yeah, no, of course, 100% code easier.

SPEAKER_07 1:26:09

Like a Mono Repo is also the way to go, of course, because if you have multiple repositories, it's much harder to coordinate your coding agents across them. Like it's possible. I do it for like one particular repo, but we have only two pretty much. Uh so otherwise you want all the code and all the context in the same place. Like we're context maximumists.

Henrik Göthberg 1:26:30

This is the whole point. This is why everything is in GitHub.

SPEAKER_00 1:26:32

Yeah, and we talk to uh you know CEOs or uh customers that basically sat like, oh, I was on a rock concert on the way home from the train, I just sat and vibe coded a replica of my site, put it on GitHub, and now it's performing way better than my CMS.

Anders Arpteg 1:26:47

But if we go a bit more philosophical here and and uh say that okay, AI is allowing us to remove some of the abstraction levels we had in the past, like a CMS system is something that non-coders can use to then build stuff and fig mice as well. And then you can move down the abstraction level and say we can work on the coding level for non-coders directly. But then if we think even further here, and I'm going a bit extreme, but I'm quoting Elon and I'm actually quoting uh Andrew Andrew Carpatha as well, saying he tried to build this um some kind of menu app or just for for for sake of experimentation. But then you can think okay, before you you you had some kind of menu you wanted to have, you could do that in a CMS, then you can switch to code, you can do it very easily. But potentially you can move down and say, I just want to use Nano Bonana or an image generator to to do the actual menu, and I don't need the code in between to do so. Because the code in itself is just you know rendering some kind of image anyway. So why not cut the code out as well?

unknown 1:27:51

Yeah.

Anders Arpteg 1:27:51

And then Elon has said you know sometimes, you know, uh he believes you know the future thing with apps and having a mobile phone and that you may not need apps at all. And the only thing you have really is an AI and some kind of visual interface where the pixels are being generated directly by the AI, and you don't have these kind of steps in between with some machine code or some Python code or TypeScript or some kind of CMS system. You just have the AI generating the pixels like nano banana style. And that's the way interfaces will work in the future. What do you think? Is that too extreme?

SPEAKER_00 1:28:25

Or I mean it's probably the interface for humans. Yeah. But then you have, of course, agent-to-agent interface, which if you're going that far into the future, then I mean I'd see a home page today, like uh you know, a rendered page, basically. But in the future, you will probably have some part that is code, which is like open APIs that the agents can use, and some part like you're describing, the images or um you know customers.

Anders Arpteg 1:28:51

The agents could talk in latent space, right? Or they they don't need pixels, right? I guess exactly. It would just burn too many tokens, I guess.

SPEAKER_07 1:28:59

Why would you need exactly right? Yeah, but on the other hand, I mean, I in I think in this example, which is interesting, right? Uh the fact that the model is supposed to take this menu and then visualize, like his app was supposed to do, is probably also some prompting or memory, right? Because he would need to ask for it the first time, or if it happens without asking, and he says, like, don't do that again. I want to actually read, like with the through my AI glasses or whatever devices he's using, right? Then it needs to remember that and don't do it. And then this like instruction, do or don't, is also kind of like code or a prompt, right?

Anders Arpteg 1:29:36

So it's English at least. I mean, it's the human kind of sensors, yeah, exactly.

SPEAKER_07 1:29:40

So the question is like, like, do you want it to be like um and but you're gonna store it, right? As some sort of context in this like whole memory for this multimodal LLM or something like that. But so then the question is like, do you want it to be probabilistic or deterministic? If you want it to be deterministic, then you probably want code because it can be repeatably verifiably done again, hopefully. Uh, but if you want it to be kind of more probabilistic, more intention level rather than like exact specifics, then you can do it more as a prompt.

Anders Arpteg 1:30:13

Can you give an example when you want it to be deterministic?

Determinism Memory And Trust

SPEAKER_07 1:30:16

Yes, absolutely, absolutely. So I have this project at home. Okay, I'm rebuilding a Volvo Valpap.

Anders Arpteg 1:30:22

So okay, okay, but please.

SPEAKER_07 1:30:23

Yeah, okay. So I have a Volvo Valp that I'm rebuilding to an electric car. So I'm gonna swap the uh petrol engine to electric. And to do this, I need a coupler between like the gearbox on the old car and the splines of the shaft of the new electric engine.

SPEAKER_04 1:30:39

Right.

SPEAKER_07 1:30:40

And this is a fairly custom part, like there are one guy that I know that has done this before on the whole internet pretty much.

Henrik Göthberg 1:30:48

With Volvo specifically, yeah, with Volvap specifically.

SPEAKER_07 1:30:51

So basically, then I need like somebody to cut me these splines for the gearbox because the motor choice is kind of common.

SPEAKER_04 1:30:58

Okay.

SPEAKER_07 1:30:59

And to test this, then I basically had Claude like I took some pictures of the splines, gave them to Claude, and had Claude produce code that will then 3D print to these uh splines, and then I can test it. But now that I tested it, I want the exact same outcome again when I go to the machine shop and cut this out in the metal. I don't want a slight variation, like the first attempt actually, like Claude made uh the splines inverted shape. So like they were a circle in the middle, they were wider at the wrong end. So after you tested it, you want it to be the same all the time. Yeah, so I love that this is determining. Yeah, so if if if I just like told the same thing again, it might make a different mistake the next time. But now I want it reproducible because I tried it in plastic and next time.

Anders Arpteg 1:31:48

Is it that just to memorize something rather than actually generate something?

SPEAKER_07 1:31:52

I mean that it would memorize the the code that you like. This is like it's a Python code that outputs a 3D model.

Anders Arpteg 1:31:59

Yeah.

SPEAKER_07 1:31:59

So should it memorize that specific code? But then it's store this code in the memory.

Anders Arpteg 1:32:03

So it you could just memorize the exact like uh 3D model that you have produced from it, right? That you don't need a code at all.

SPEAKER_07 1:32:11

Yeah, yeah, potentially.

Anders Arpteg 1:32:12

But then like that's also like uh I suppose uh I actually had this kind of similar discussions in the past, and uh you know it's it's actually I think it's super hard to really motivate really really strong determinism here. But I could be wrong.

Henrik Göthberg 1:32:29

So I think I think it's a deeper question here because you you are speaking on this on such a high level or nerdy level, and when and when you hear the fundamental media or someone who's not into the details, I say, Oh, how can we have um uh stochastic systems ever do reproducible tasks? And I say, Wait, well, you are not calling LLM the system. I hope. I hope you have a harness. I hope you have uh a compound system, don't you? What do you mean? Like, oh my god. I mean like so so we're literally trying to use math that is not done for reproducible things. So it's so what what I'm hearing is like how we look at the whole to where where do we want the uh how do we want to use it, and then when then then ah, this is actually memory. I'm from memory. So then you worked around the deterministic problem, right? So this is what your angle is that yes, we need reproducibility and exactness, but we can get it in a different in a different way, or what is your angle on this?

Anders Arpteg 1:33:25

Well, some people argue from a security point of view it's super important to have deterministic uh solutions that you can also have transparency in, for example, and also that reproduce the same thing.

SPEAKER_07 1:33:36

And verifiability in my case, right?

Anders Arpteg 1:33:38

Yes. Uh but you know verifiability I think you can solve by by having good memory instead.

SPEAKER_07 1:33:45

Yeah, yeah, potentially if you memorize it in the same form, but then maybe the like the most uh dense way to save this is actually the program that generates the 3D model.

Anders Arpteg 1:33:56

Could be, could be. Yeah, maybe, maybe not. But anyway, I I think it's an interesting question. Agree to disagree. I think you know, so many people are stuck in the idea that you need to understand code to trust anything. And that is the problem I'm having. So if you simply believe that just because you have the Python code available somewhere in some repo, then it's trustworthy. I don't buy that argument because code you can't really trust if you have a million lines of code, how can you trust that? You can't as a human have a full overview of everything happening there, especially in all Python code, since there's like uh infinite layers below, pretty much. I think people have an overbelief in the trustworthiness of code compared to models. And that's what I'm arguing with.

SPEAKER_00 1:34:45

Okay. Do you wipe code by the way?

Anders Arpteg 1:34:47

Yeah, reluctantly. Reluctantly, yeah, I can hear that. But I'm getting But you read everything that it's No, but I can see many cases where AI do not solve the task and I have to help it. I mean, if we take a simple example of like some kind of bug that is a bit more difficult than a standard bug, as a human, you can actually put in a breakpoint and actually do troubleshooting in a very interactive way, which an AI actually cannot do today.

Patrick Couch 1:35:15

Yeah.

SPEAKER_07 1:35:16

Or can it? I mean, depends if you use a CLI tool to do it or not.

Anders Arpteg 1:35:20

Well, they can put like print statements and stuff and run it and see what happens. But we can as humans do a much more, I would say, efficient interaction in the debugging phase than AI, at least today, can, until they have you know that type of functionality.

SPEAKER_07 1:35:35

Yeah, but like if if you take like a C debugger or something like that, like there are interactive interfaces, but if they would break down that into like a CLI with sessions.

Anders Arpteg 1:35:42

Yes, but they don't do that today, right?

SPEAKER_07 1:35:44

No, of course. But like they could if the future spend a lot of time like building kernel code, for example, that's what I would do. Like I would make sure that the agent could do all the design. Yeah, and some people probably do, but like I mean it's very unevenly distributed for sure.

Anders Arpteg 1:36:00

It would it would it will count for sure because I think it's so powerful. And I can see myself when I fix some bugs that uh I try to use the agent, I always do, but then it fails, and then I go in and I can then so much easier to fix it. So there are many cases uh just by reading the code, yes.

SPEAKER_07 1:36:16

Okay, but I mean there are examples like this. For example, like this whole Chrome debugging protocol where that basically gives you this Chrome inspector that web developers use to fix pixel errors, like we will use that already, like through an MCP. And it's super super powerful. The only problem is that it's also super super slow because it enters this like debug mode of Chrome. So I don't use it all the time, but if there is like a pixel error that I can't solve by just prompting and prompting again, then I'm like, okay, go in, check all the CSS, all the HTML, like I can't bother to do it, like you solve it, and it solves it.

Anders Arpteg 1:36:51

So and and I'm sure that you know Claude and um Kurser and others will have debugging proper you know functionality properly soon as well, and uh they will unless they will have computer use and CPUs keyboard and mouse, unfortunately. But uh, of course, that's where where most models are really trying to achieve you know good results in. And and GPT 5.5 that was just released, I mean they are really performing well in computer use. Yeah, yeah. A lot is happening there.

Henrik Göthberg 1:37:14

But it's but then we can keep the CMS because the models are going in the agents using WordPress. Yeah, exactly. Great back to square one, which way to go?

Where Humans Still Beat AI

SPEAKER_07 1:37:24

Yeah, exactly. But like we had this thesis like back back in early, early 2025. This was like the first uh computer use build that was released by Anthropic, I think, autumn 24. And we're like, oh shit, this is amazing. No, we don't have to build MCPs, we can just like let it loose. And we we tried, uh we built our own, like a copied of their sample project for like spinning up your VM, uh opening the browser, letting loose the computer use, and I was gonna do a demo to Tom and like look here when it posts in WordPress on our own site, and it's just missed the fact that it was still in the title field and just typed out the whole blog post in the title. I was like, oh.

Anders Arpteg 1:38:02

But I think people you know don't understand that AI, in some ways, are actually very stupid. Um if I just can elaborate a bit on that. I mean, uh I I usually use this kind of pyramid from OpenAI to explain my view on AI, and it's extremely good in memory, meaning you can actually put 10 books into the prompt and it can have more or less perfect recall in all these 10 books in a single prompt. Like that no human have any way and any chance to do anything similar to that. But then if you take something like ArcHEI3, the recent kind of benchmark that came out, then you need to be able to interact with some kind of interface and you need to reason about it and take action. And it performs horribly. For humans, you get 100% accuracy, but any human put in that test can solve these kind of tasks, and that's the state-of-the-art AI get less than 1% accuracy. And and they they don't really have the ability to interact in an interactive way, um, and and then reason about that in the way that humans can.

SPEAKER_07 1:39:01

Yeah, and they have no training in that modality pretty much.

Anders Arpteg 1:39:04

Well, they are partly. I mean, computer use is that partly, isn't it?

SPEAKER_07 1:39:08

Yeah, in a sense, but it's still probably like screenshot-based and like pretty slow in the sense where like I mean Gemini can watch videos now as well, so it's getting a lot better.

Anders Arpteg 1:39:18

But but you can say that it it's not trained that way, but in reality it doesn't have that capability.

SPEAKER_07 1:39:25

And it's because it's not trained that way. But once they do, like it will it will ace that.

Anders Arpteg 1:39:30

I don't everyone is trying to beat Orc AGI3, no one has done it. Yeah, but it's been out for what? Less than a year. But everyone is trying to solve computer use for ages as well. I mean it's it's super hard. Uh I wait to wait until 2028, I think. Then I think you know it's gonna be hard. I mean it's harder than people think.

SPEAKER_00 1:39:47

Uh I I would really argue that. And uh but why do we need computer use so bad?

SPEAKER_07 1:39:52

Uh one reason it's it's called like uh banks. They don't have MCPs, and like you still need to go in and find boring receipts and stuff. So like you just use computer use for that. It's amazing.

Anders Arpteg 1:40:05

We can go into the whole Jetpa and uh Jan Likun argument here, but that's the reason I think you know we need to do some changes to actually get this type of intelligence. I don't think we're there yet. And I think there are a number of ways that we need to step uh or take to actually achieve this.

SPEAKER_07 1:40:20

And you think that goes like beyond the transformer models can ability.

Anders Arpteg 1:40:23

No, it's still based on transformer, but you do latent, you know, latent space reasoning then in a different way. We actually seen a number of papers going that route, so it's starting to move in that direction, and of course, all all the image generator is not reasoning in pixel space. They are first having an autoencoder around the image and then doing the diffusion with transformers, you know, in latent space. So we already have any kind of you know model that's working without text, it needs to be in latent space. So I think we are seeing clear signs in moving into latent space reasoning, which Jeppa is uh is primarily about. And we need to take that to I think the next level to have this type of reasoning which is not auto-regressive, um, uh, to make this happen. And I think we will count.

SPEAKER_07 1:41:10

And isn't that basically like the innovation of JEPA then that you do reasoning in latent space rather than just like uh tag on after the fact pretty much with tokens?

Anders Arpteg 1:41:19

It's a big yeah, I mean the joint embedding is really about that. So joint embedding prediction architecture is really, you know, taking both X and Y, put it in the embedding space, which is the latent space, and then you do the prediction that are not the autographic one. But but once you have that and the world models that everyone is speaking about right now, where you can actually actually judge if a position is good or bad. Uh if that's for physical AI, where you try to understand if uh if the gravity is working in a certain way or whatever it is, uh that can that can be something. But if you have this kind of, you know, they have like four or five different modules that needs to work together to achieve this. But if you have that together and and then do the reasoning in the embedding space or latent space, or as Elon call it the vector space, then I think we're we'll get somewhere.

SPEAKER_07 1:42:03

But it's interesting then because I mean obviously Jan Le Kun was like head of AI at Menta for quite some time and like should have had a lot of resources. So, like, is it that much harder to do that? Like they tried and tried and tried, and then they did like llama on the side and it failed, or like well, I mean because I I kind of disregarded him because it just sounded like a grumpy old man, pretty much like complaining on all the other labs.

Anders Arpteg 1:42:26

Yeah, no, but I think you know he has failed, and that's a bit sad. We'll see if they succeed with a new AMI lab that he's having. But I think you know, with all the papers and even all the other models that moves a bit beyond text, because text is really latent in some way, the level of abstraction that text has is rather high compared to pixels and audio. Right. Audio and and and images is super high in dimensionality, so then it's super hard to reason in that space. So to be out-regressive in pixel or audio space would be idiotic, and no image model today is doing that. So they already have moved to that space. Right. So it's already taken that kind of step.

Henrik Göthberg 1:43:06

But you can also, I mean, like we had Karim here, the founder of uh what's his super intelligence something super, you know. So so startup in Sweden that is more working towards um, you know, they are they are saying uh we we believe in a different route towards AGI, which is more or less more carefully understanding, well, going language and text as a way to AGI, hmm, will that really work? Do we have the real uh real real um understanding of how we build stuff? So they're rather going the physical route. How do we build a physical world model for a for how do we learn to simply grip something and move something in a warehouse? And that is sort of the the baby step of something that builds a world model that becomes more and more effective. And we're basically there is some researchers in Sweden which basically says, hmm, maybe not the text ways the right way. And and and I think but but here we put the emphasis even stronger on the world model, the memory, and the recursiveness.

Anders Arpteg 1:44:07

But just to summarize it, I think the important point here is to still understand that humans are better than AI in some tasks. Very clearly demonstrated by ArcHI3. I mean, yeah, for sure. So easy for humans to do this task and AI, independently of why, it still can't do it today, right? So there are things that AI cannot do today, and humans do much easier than AI, but certainly the opposite as well.

Henrik Göthberg 1:44:31

So if you have that understanding, I think you have a much bigger chance of like butt and the core question is also do we simply need to do more of the same we're doing, or are there fundamental architectural tweaks missing? And I think that's I think something is missing. This is my view.

Anders Arpteg 1:44:48

Who knows?

SPEAKER_07 1:44:49

Who knows? Agreed, it could be, but I think you told me this, Tom, from from the white combinator event yesterday that like actually AI doesn't have a problem, so there's still like a space left for humans to have like the problem that AI can fulfill. And that's where like if AI is gonna take all jobs, like the humans are gonna have problems, and that will kind of generate new work in itself. So there is this like human agency aspect that that is like uh I think at least for the time being, like uniquely human, and and obviously, like what ARC AGI free is kind of edging towards is maybe like part of the same capability set.

Anders Arpteg 1:45:31

I mean, just learning so quickly as humans do from a few examples and then extrapolating from that is something that today's models have a really hard time doing. Yeah, and uh but it will come there, of course.

Henrik Göthberg 1:45:43

It will come, but uh the the there's some breakthroughs left, yeah. Some small ones maybe.

Anders Arpteg 1:45:50

Who knows? But um it won't be that long. I don't know.

Henrik Göthberg 1:45:54

I don't know. That's another question. Moving if we move truly philosophically.

Anders Arpteg 1:45:57

How long before Arca HGI3 will be resolved? Uh I guess one year. I don't know. Yeah, probably.

Henrik Göthberg 1:46:03

So the crazy thing, we've had a couple of benchmarks, oh, it's impossible, and then three months later is oh shit, we're on the we're on the move now. Yeah, yeah.

Agentic Marketing And Personalization

Anders Arpteg 1:46:11

So we have seen that one. Cool. Uh let's see. I'll skipping some topics here. Time is flying by here. Um yeah, so okay. So we have more complicated potentially marketing tactics happening. Um what do you think? You know, how what will the impact of AI agents you know being put in in a more agentic way be on you know, on the tactics of marketing be in the future? Um, do you see what I mean?

SPEAKER_00 1:46:42

It's a bit complicated. Yeah, I think I mean when we say marketing, marketing is a very broad word. I mean, we have lots of like branding and creativity that I think is still very human domains. Um where in our space, which is more kind of data-driven marketing, that's where an agentic uh kind of setup works really well because it's all about analyzing data, uh, you know, adjusting what's going on, running experiments, and then create that loop. But I think like uh Legora did with Jude Law, did you see that? So they basically have Ju Law as their brand ambassador, which was like, holy shit. I mean, Jude Law is one of my favorite actors, and I never thought he would go out and basically sell himself, especially like you know, Hollywood with AI, that's like it's not the best combination. Um, but in the end he decided to do it, and I think that's uh you know, truly uh Klarna did it with Snoop Dogg as well. You know, that type of marketing, I don't think an AI would be like, oh, spend uh I don't know how much Jude Law take, but it's probably not cheap. Yeah, like spend a couple of million dollars on Jude Law to get attention.

SPEAKER_07 1:47:56

But but I think the flip argument to that is like I mean, if you listen to biolabs and stuff like that, they say like we try millions or billions of molecules now with AI. And I mean Meta recently published uh a prediction model that will predict uh the emotional response in humans based on the video. They've done like FMRS. So you just do the same setup, right? You generate the billions of potential ads, check the emotions of the potential simulated users that you want to reach, and you're gonna find your law or something better, I think. With AIDS compute and the agents to work.

SPEAKER_00 1:48:30

I mean, at least if if you were to listen to the latest years of like influencer marketing, everything should be authentic, you know, even big companies like Wattenfall or whatever, they shouldn't spend millions on uh commercials, they should have a handheld mobile because it's authentic and that's what people relate to, and things like that. So, I mean, of course, you will probably be able to, and uh Meta also has have this vibe uh social network where they generate all the content with AI. So if you kind of combine that with uh um brain scan models, then you have a pretty scary uh setup where you can basically decide what reaction you want from your I think Elon puts this as like AI is going to one-shot the human limping system, and I think he's right.

SPEAKER_07 1:49:13

Like, I wouldn't bet against Elon against Elon on that point.

Anders Arpteg 1:49:18

But speaking in general, let's take the the Swedish election now coming up here. You can imagine a lot of marketing being done here for that, of course. We could even see, I think um the election in was it Venezuela? No, Argentina, Argentina in uh two years ago or something. I think they call it the first like AI election or something. And they used a lot of generative AI to put up like bad pictures of the but they were do you think it's fun because they were road roasting each other.

Henrik Göthberg 1:49:45

So it was people, it was obvious it was fake, but it was still uh communication wise, it was quite powerful.

Anders Arpteg 1:49:51

And of course, we're seeing the same now with Trump and one of the other kind of weird uh AI generated images, being actually not trying to make it look real. Just using it as a communication tool, I guess.

SPEAKER_07 1:50:02

Yeah, and in the Iran war, like they have been pretty aggressive, haven't they, with like the Lego Trump and like the lot of people Trump waiting for the Iranian uh delegation, yeah, like uh fake, obviously fake videos, and like they still use that to try to get their message out.

Henrik Göthberg 1:50:17

So this point is that it's completely AI generated, but it's not trying to fool us, it's trying to use it as a powerful communication persuader. Yeah, but I think it's very hard to draw that line, isn't it?

SPEAKER_00 1:50:27

Of course, I mean I've I've seen some on LinkedIn where they even write on the uh image itself, like AI generated, but people are still like blowing up in the comments and be like, oh, this is a Palantir CEO with uh you know our uh state uh head of state, you know, and so I think that that line is hard to draw.

Anders Arpteg 1:50:48

But if we just try to be really uh visionary here in terms of what marketing will look like in the future, one one of course is simply to use AI and agency to uh to to optimize the the marketing that you have, yeah. But I guess you can also think of the product that you you market to be more AI driven, meaning just as Spotify could have Discovery Weekly, which is personalized to hundreds of millions of users, and then adapt it to this is actually the music you'd like. Yeah, perhaps we can have ads in the future that is so personalized and actually having this kind of they understand you as a human so much, so they actually have different videos that is personalized.

SPEAKER_07 1:51:29

Like one example that we've seen recently, like with our our growth agency, growth hackers, is that we started running Cogni. Yeah, Cogni does a lot of analysis of the data, yeah, and all of a sudden it's like you should really like uh build uh some SEO page for Mark Nasfering's bureau, like marketing agency. We're like, we're not a marketing agency, we're a growth agency, you know. And and this hadn't crossed our mind really, but we were buying it on search, of course, because like people looking for our services think of us as a marketing agency, right? So, like from a brand positioning perspective, like it didn't make sense as our identity, but it made sense like how people uh tried to find us anyway, because but they didn't know about us, maybe. But then if we popped up, it was an interesting proposition. So that's kind of working our way into more of the idea space, the feeling space, the like what what people uh yeah, how they reason about this. So I in that sense, like AI can already help kind of take away our prejudice about who we are and try to position us better in the market.

SPEAKER_00 1:52:36

So I think also like historically, uh you've had a page or a landing page, and and you you try to run A-B test on that in order to improve conversion rate and things like that. But and and if you're even a little bit more advanced, you start to do personalization, which means parts of the sites adapt depending on who you are. And I think AI, in a sense, completely will rewrite that because why would you need one single user journey? You could have like Spotify for hard rock players, and then the the that user journey can specifically for people that like to listen to hard rock.

Henrik Göthberg 1:53:09

You can have someone for blues, you can have like all the subcategories, you know, AI can generate millions of user journeys that are specifically for but I think that's what you're also aiming at then, because then marketing and product is converging in some ways.

SPEAKER_00 1:53:24

Yeah, definitely. I mean, I would say that it depends, like in in our industry, in the startup space. I think I mean if you are at Spotify and do marketing, or if you are at Lovable, you know, yeah, what are you doing? I mean, it's a great product, it sells itself, right? You you just try to amplify that. Uh, and I think that's a big difference than if you're at uh I don't know, an energy company that but I think it's also from the from the loop, right?

Henrik Göthberg 1:53:50

So what is a product, right? And ultimately, if you have fundamentally personalized perceptions based or journeys, yeah, then the meta product is fundamentally different products.

SPEAKER_00 1:54:00

Yeah.

Henrik Göthberg 1:54:01

I mean so in this sense, then you have a marketing intelligence product loop that is sort of I mean it's the same value that you give, yeah, yeah, it's just adapted to who you are.

SPEAKER_00 1:54:11

Yeah. Because I I mean it's if Spotify is still a music service, it serves any music lover. But if I feel that I'm at home as a hard rock lover and the user journey is adapted to that, then I'm gonna feel more connection to the brand. And that basically makes the brand be a good idea.

Henrik Göthberg 1:54:25

But it makes total sense then because if if we are talking about product as something that you employ to do something that solves friction or something you want done, so I want to employ a product that is perfect for my use, and all of a sudden, then what where is the line between what is product, what is marketing, what is UX? It's it's just it commercial.

SPEAKER_07 1:54:49

Maybe more of that moves into your personal AI agent and like product companies left is doing more like uh agent-based payments, APIs, information that's really clear.

Henrik Göthberg 1:55:00

Because marketing is then your meta product, or the the way you experience the product is part of the meta product in that sense.

SPEAKER_00 1:55:06

But how you explain it to different people, I mean the the value and the product in itself is what solves the problem, right?

Henrik Göthberg 1:55:12

Yeah, so the core the core functionality might be the same, yeah. But the the but and and maybe that's an inner core end. But I'm not I'm not I'm not layering and layering layering until like you know, where where does product stop and where does marketing or or experience start?

SPEAKER_00 1:55:27

I mean that's uh that's an interesting I think it's converging.

Henrik Göthberg 1:55:31

I think it's converging.

SPEAKER_00 1:55:32

That's my I think like uh you you know when you get to know a friend or anything, you you want some common interests and things like that. But at the basics, it needs to be a strong relationship. I mean, it's kind of the same that with the product, you have to love the problem. You have to have a problem with the problem that it solves basically, but then it helps if it's closer to you, it looks kind of the same to you and it it talks the way that you talk, then you come a little bit closer, which means that you can you will pay more for that than uh an exact replica of that service that just looks but let I let I just want to trust uh close this and go even further.

Henrik Göthberg 1:56:09

If I put the thesis down, marketing will never be further away from product than it is right now. It will just converge and get closer and closer and closer in terms of how this you know, we always talk about product, marketing, and then we have the distance between marketing and sales. I mean, like so we are talking about really that distance goes away, and I'm saying, well, it's not gonna go away, it's just gonna go closer and closer.

SPEAKER_07 1:56:34

It's the context convergence. Yeah, isn't it?

SPEAKER_08 1:56:37

Yeah, it is.

Anders Arpteg 1:56:38

But I think I think so too. I mean, uh as uh ads expert as well, uh that you are in from the past, you know, if we do consider, of course, Google search going down in popularity, it's just agents you know looking at Google search in the future potentially, and then people are looking more and more at the yeah, text output or video output or yeah, uh image output output. I mean, uh OpenI wanted to move ads into their output, so to speak. What do you think? What is will be the proper way to just you know get the message out there, so to speak, when you can't show ads in the traditional way?

Ads In LLMs And Funnel Shifts

SPEAKER_07 1:57:17

So there are multiple ways, right, to show ads as well. I mean, Google has had a super strong position because they have a way to show ads where people have already have they already have the intent, they are at the bottom of the funnel, they figure out I need now to book a concert ticket, blah blah blah, and they Google that and they find it and they're happy. But what happened in the moment before, like we call this the upper funnel or awareness stage, for example. Uh that is of course still left, even if the agents do like figuring out how to help me solve my problem, but but creating the problem that's still marketing as well.

Henrik Göthberg 1:57:52

Upper funnel, yeah, upper middle bottom, yeah. Or associating yourself with the problem area, exactly.

SPEAKER_07 1:57:59

I mean and that might become even more like a creative, uh exploded by AI basically. Like, I mean, this due law example is is a good example, I think. That they're basically like they're borrowing a little bit of his brand to also amplify their own. But like maybe there's a crowdsourced version of this. Like, I figure out that I need everybody to eat my ice cream, so I like give it to certain people. I have AI help me find this, and then like by them eating this ice cream, it now becomes valuable in the eyes of others, and now I've done marketing, and it's like agents then get hooked in to find the ice cream and buy it, right?

Henrik Göthberg 1:58:39

I like I like the way you painted the funnel now upper, middle, lower, and how we can understand how agents are helping us in different ways depending on you know what do you have a lead generation problem, you have a lead nurturing problem, you have a closing problem. It makes total sense to understand then that we need to categorize different types of marketing workflows in relation to creating the demand versus closing it.

SPEAKER_00 1:59:02

Yeah, but I think the uh the like where you start with an LM, that's not fully understood. Like, do you start in the awareness stage to be like, oh, list all telephones? Exactly. Or are you just asking like what's the best energy provider in Stockholm?

Henrik Göthberg 1:59:16

Well, like I I I I if you like watches, it starts with you nerding around watches, going on YouTube, doing different things, and all of a sudden, oh, I like pilot watches, oh la la la la, learning about that, and then all that's my nurturing story already there, right? Exactly top or funnel to bottom.

SPEAKER_00 1:59:30

Yeah, but I think in uh historically what we've done is we've of course mapped all the clicks that happens to a page, and then we can see like, oh, this user had uh, you know, within the 90 days that we know that it takes to consider a new uh telecom provider, for example, we can see that oh, they started searching here, they started looking at which type of uh abandoning or like uh subscription they want, and then like 90 days later they've ended up at what they want, and they can kind of uh choose that. And then we can map out all those clicks and understand like these clicks are top of funnel, and then we can do what's called attribution, we can attribute the value all the way back. But I mean the LLMs kind of destroy this that's a different story. The whole that whole exploration is done in the LLM basically, and there's not a single like call that goes out to the page.

Henrik Göthberg 2:00:17

But isn't this where the power shifts then? Because somewhere in my chat repository, uh Chat GPT knows exactly which uh toys I'm dreaming of. You know what I mean?

SPEAKER_00 2:00:29

Yeah, but I guess like how can he how can he monetize the difference is like if you present, I mean I heard someone compare like Google Ads with magazines, because basically if you look through an expensive magazine, the ads are as good as the content. Yeah, and that's true for a Google search as well. I mean, the the ads are as good as the uh organic content. But if if you uh ask LLM a simple question, it can't give you some like, oh, this is kind of good things, but there's also this. I mean, they have to give you an answer, uh probably a pretty deterministic answer. Yeah, they can't give you a lot of things.

Henrik Göthberg 2:01:03

Otherwise, I'm not kind of happy with his exactly.

SPEAKER_00 2:01:05

So it's kind of hard to fold in ads into the format in one sense.

Henrik Göthberg 2:01:10

Yeah, so that's an interesting one. Where where will that go? Or will it will it mean that the the the like the normal search and ad space is not going away? What what what is the prediction here? Because you you're you're pointing at something spot on.

SPEAKER_00 2:01:23

Yeah, exactly. I mean, um Meta has this model, of course, that's trained on the three billion users that they have, so they're supposed to know you. Um and I think that if you have a model like that, then that model can interact with different MCPs or different interface, whatever it is, and then you can kind of sort between those and then get a couple of responses back. So basically it's a data exchange between the advertisers and the LLM.

Henrik Göthberg 2:01:52

But you can't present that in the will ChatGPT ever sell the top right corner to a back to banners in ChatGPT, right?

SPEAKER_07 2:02:04

Yeah, but maybe I mean agentic commerce is an obvious way to try to bring that monetization in there in the like a genuine way, right?

Henrik Göthberg 2:02:12

Do you I mean like prompting? You know, you know when you're prompting, and then now I could do this, this, this, this, even even in the prompting uh you know structure, he could easily slip in. Do you want me to uh search where I can find this? If that even happens right now, it already happens right now. Do you want me to find which uh which places you can buy this? We are doing it already in some ways. Yeah, it is happening.

SPEAKER_00 2:02:33

I mean if you ask you, you're gonna get an answer if you ask in a lamb which uh telecom provider you should have, or you know, but of course, also like discounts could be a way to do it.

SPEAKER_07 2:02:41

Like you ask for five Mexican restaurants, you get five, but then it's like do you make the Tex Mix uh restaurant like with a little bit of Texan flavor with a five percent discount, and you're like, Yeah, maybe that's fine. Like I can do nachos with steak or where will this go?

SPEAKER_00 2:02:56

Yeah, but in the end, you I guess you're nudging the trust of the LLM if you start showing ads. And I think that's the whole point of Anthropics whole uh commercial series that we're never gonna do ads because it jeopardizes the whole trust of the platform basically.

Anders Arpteg 2:03:12

No is it I don't know if you have to pay for it in the future, perhaps people will still accept it, you know, if they get it for free. Yeah, yeah.

SPEAKER_07 2:03:21

No, I mean it's the whole also like B2B, B2B, B2C place that like anthropic, heavily enterprise focused, while like open AI has gone like really broad, a much more adopted free plan, and like maybe ads is the only solution that they can provide this in the future with. I mean they tried this like uh scan your eye for a crypto approach as well, but I guess that backfired them yeah, a lot will change for sure.

Anders Arpteg 2:03:48

Uh and uh yeah, uh I mean marketing is is still a very important tool to have, right?

SPEAKER_07 2:03:53

I mean, otherwise we can't grow businesses, and uh we need to figure out a way to do this in uh in a genetic future in some way or for yeah, and if especially if like for example SAS is becoming like completely commoditized, or you can even like just spin up your customer stuff. Like, I mean, I think we're looking at a new super cycle of sales in a sense. Like we have all the incumbent companies, they sell their products, they have their stock valuations and all of that, but like AI factories is coming, humanoid robots is coming, AI capabilities is coming. So, like everybody that's not uh like a pure play AI company is gonna get squeezed, pretty much. And then they need to do more selling and more ads.

Anders Arpteg 2:04:35

Okay, I'm just trying to think if we go even further, do these kind of extreme use case thinking like Elon often does, thinking that okay, in the future we won't have ad, we won't have apps, we just have a graphical interface or a video being streamed to us all the time by an AI and then understanding what we're saying and and doing all the time. We still want to have you know companies that want to influence us and inform us of their products or the services somehow, still want the illusion of choice.

SPEAKER_07 2:05:09

Yes. Unless you're talking about like the post-capitalism world, I mean that world of abundance that might eventually happen as well. And then of course the rules of the game are different, I think.

Anders Arpteg 2:05:22

Speaking of the world of abundance, um, if we were to move to an AGI future, we mentioned some years here, perhaps 2029 uh or something, or or do you have what do you think, Anderson? Well, I've been saying Ray Quer 12 the year for yeah. Yeah, you copied him.

Henrik Göthberg 2:05:39

Okay, I can I can concur. He copied the Ray on the first episode, I think.

SPEAKER_07 2:05:45

It feels like the safest bet, right?

Henrik Göthberg 2:05:47

No, it's interesting how it sort of did we uh I I don't know. Yeah, it's the safest bet.

Anders Arpteg 2:05:54

I think it actually makes a lot of sense as well. Uh I mean a lot of people are saying one year or something, I and some people are saying we have it here, and I certainly don't I mean it's so easy to show, just it can't do ARC AGI 3. I mean, obviously we don't have it here today.

SPEAKER_07 2:06:06

But um but of course, like from uh uh commercial intents and purposes, like I don't earn money by solving ARC AGI, but like No, no, but there are things AI can't do.

Anders Arpteg 2:06:17

So if you define AGI as something that can do on the part of an average human co-worker, we know it's a lot of things it can do today, right?

SPEAKER_07 2:06:26

Yeah, yeah, and at the same time it can unlock like tremendous value that like a human can't. So like it depends like which lens you're looking at is from. Like, I mean, from from the SAT world or whatever, like maybe it's already here, but obviously from the like uh I want to help with the finding out the meaning of life world, like maybe it's not.

Henrik Göthberg 2:06:47

But then and then you have of course uh okay, we can do now AGI uh desktop. And then what about physical AGI? Uh and what about then the regulatory landscape? So we have AGI at scale, right? So so when we say 2029, which one are you thinking about then? The first level or the physical level or the abundant level? I I think 29, maybe I can agree with that, but I'm thinking the first level of sort of yeah, I'm thinking digital space. And and I'm not saying physical space at scale. I don't uh because I don't think the regulatory will be there.

Anders Arpteg 2:07:26

But it could actually be, I think the physical space is moving very, very quickly as well. Yeah, yeah. Yeah, for sure.

SPEAKER_07 2:07:31

Like faster than humans, obviously.

Anders Arpteg 2:07:35

Yeah, uh I will see.

Henrik Göthberg 2:07:36

So it's not impossible both can happen in 2018. But are you are you on that sort of trajectory or 20 years or 50 years?

SPEAKER_07 2:07:42

I I think my timelines are like really compressed. Like I I I usually maybe my timelines are too compressed that I expect more to happen than those.

SPEAKER_00 2:07:51

Yeah, we have this problem that we usually are too far ahead, and we sit and talk to people and they're like ah we don't understand what you're talking about, and then eight months later there's another company doing the exact same thing and they're having success. So I mean uh market window.

Henrik Göthberg 2:08:06

Yeah, exactly.

SPEAKER_07 2:08:07

But but I think I mean I'm I'm worried that like uh do we really discuss what uh post like AGI level one, if you like, then world looks like in the this upcoming election in Sweden, for example. That's a good point, and I think we should like because there should be like a democratic mandate for the plan. Like, is it just are we just gonna all leave off the like per year uh find the head, like this big rare earth deposit that they found outside of Kiruna? Like because I mean a lot of the Swedish uh budget that goes to the state is based on labor taxes, right?

Anders Arpteg 2:08:45

But I think it was also important to distinguish between the you know the technical progress of AI, which will go probably very, very fast, and that that is what I mean with 29, meaning that we have the technological possibility and potential of AI, but then that the actual you know companies will adapt, the society will adapt and much, much longer than that. I agree.

SPEAKER_00 2:09:05

I think that when when when we talk to clients now, it's more about them adapting to AI, it's not about the latest models or like it's it's about them basically. Um, and it's interesting because some people, of course, lean more into it and others are more hesitant.

Anders Arpteg 2:09:18

And I'm really scared about, and I wish politicians spoke about that more, is the concentration of power and the AI divide now being created, because that will go just faster and faster. Yeah, some companies that really do make the proper use of AI they will just accelerate insanely faster. Yeah, that's very scary to me.

SPEAKER_07 2:09:37

Yeah, yeah, and it's it's I mean it's putting a lot of companies or like livelihoods or like uh revenue or BMP under exposure. I mean, for example, all these like dark factories that they talk about in China, like I mean, they will be able to produce a bigger and bigger share of what we consume, and if they do it at like AI and robot prices, which like I guess they almost do that already, like with suppressed pages and etc. But like consumers like cheap things, and it's gonna be harder and harder if you haven't adopted your processes to AI. And and obviously, like Europe has this pretty interesting situation with like no real big super successful foundational lab, and also like having given away a lot of the robotics to China, I think.

SPEAKER_00 2:10:23

But I think like the back to the income gap because that's already a problem. Uh, I mean I met uh was that after part, as Durihoe met some politicians, um, shouldn't say which which uh party, but uh and they were talking about this, you know, what's going on in New York, and you know, we we can already see this like grassroots happening without AI. And of course, AI is going to amplify this a lot. So, I mean, I think the politicians so should start some type of universal basic income. I know Finland has done some testing around it. I mean, why not just start to test so that when or if it happens, but we need data, we need to have an understanding for what will happen.

Henrik Göthberg 2:11:08

How can we really understand that if we haven't even experimented?

SPEAKER_07 2:11:11

Or at least have that debate because maybe uh people want to vote on like let's turn everything into UBI and like skip all the basic services or the the other way around, like zero income, you just go and like uh you get like a moot book or like some some kind of stamp card where you consume using that and no money, like but back to back to I I haven't thought about that.

Henrik Göthberg 2:11:32

But if AGI technically comes within the next election period, yeah. Which we are saying then if we are following Ray uh Ray, what are the things we should be discussing then in this election?

SPEAKER_00 2:11:45

I think the UBI definitely. The GBI? Yeah. I mean we need to get some type of program like that going anyway. But we need to debate what's the options, right? So we start it. I mean, I I believe in experimentation, right? So I I think we should start testing. Um and just figure out how it works.

Henrik Göthberg 2:12:04

I mean it's so far away, but at the same time, no, you said it this it's within this election period.

Anders Arpteg 2:12:09

Yeah, it's the supersonic tsunami coming in, but but it's also adoption is later. I mean, yeah, it really affects you know what we're seeing in society will probably take a longer time. But but you can't know that. No, we can't know that.

SPEAKER_07 2:12:20

You think that you think that based on past changes, etc. But like if we think about the slow timelines, like I think it was around like pre-World War II, like there were very many horses and not so many cars. And then that war happened, and like in the 50s there weren't almost no horses. And of course, this was not with like uh online 24-7 intelligence, like a country of geniuses in a data center type level execution. But if you bundle that with like self-replicating robots and stuff, like it could go really, really, really, really fast.

Anders Arpteg 2:12:55

And one point we can be rather sure about is that some companies will move much faster than others.

SPEAKER_07 2:13:00

Definitely they'll try to build you know data centers in space and puts on the moon, like somebody will do that.

Anders Arpteg 2:13:06

But imagine imagine you have a company, just uh hypothetically, uh, that has income that is on 70% of the world, yeah, or like 100%, like you know, most thinks that the the total global economy on a thousand X or something like that, yeah, and most of that growth will happen in space.

SPEAKER_07 2:13:25

And if that's equal to space X, I mean we're close to 99 or worse.

Anders Arpteg 2:13:29

But let's say 50% just didn't make it too true. Even 50%, if one company owns 50% of the like turnover that we're seeing in the world, yeah, what would that mean? I mean, they will be much more powerful than any kind of country that we have.

SPEAKER_00 2:13:42

I think we're already there at some point.

Anders Arpteg 2:13:44

I mean, right, we are, right?

Henrik Göthberg 2:13:46

But but and and now comes why this debate is important, regardless of exactly how far we are. It goes back to the William Gibson quotes. The future is already here, it's just unevenly distributed. And if you follow that trajectory, if we think we're gonna go closer to AGI, technically, it means some companies will from now to here reap a lot more benefits. So this this this gap, this distribution of the future will be there. Therefore, how will we as a society cope with the people on the different sides of this? So the whole UBI income discussion is also about how do we redistribute uh balances? Yeah, that becomes a qu a question we can have now.

SPEAKER_07 2:14:28

Like which which balances are left as well. Like, I mean, imagine that that we had zero tax income year 2030. Like, what do we do then? Maybe we tax, like we're currently giving away mines, for example, the Swedish politics for like if you want to establish a new uranium mine in Jemtland is a big local discussion, all right? Where does those profit go? And currently, like uh English companies or whoever comes here and builds the mines, they can reap a lot of those uh profits. But in like a world where raw materials and electricity determines outcomes pretty much, which is like I guess what uh robot factories and everything is, then we need that to look a lot different, I think, if we want to have prosperity in Sweden.

Henrik Göthberg 2:15:15

So these are very valid conversations to be had now, I think.

SPEAKER_00 2:15:19

Yeah, yeah, but I think like because we see to politicians to make these decisions, but they haven't made decisions for quite some time now. I mean, if we look at the the data center build out of 930 billion dollars, who and and compare that to the Apollo project or the railway project, which dwarfs all of them, who founded that? Did the government fund that? No, no, no, it was private companies, so we're already there, I think, and and this has been so for quite some time.

Henrik Göthberg 2:15:48

So but then it then the world will not get uh balanced because then it's then it will always be the the richest is the capital.

SPEAKER_00 2:15:57

Yeah, I think we're we're all oligarchs in one way, you know, runs the world. And and unfortunately, I mean in and we do see some um maybe hopeful sparks uh that politicians can start to take back some of that power because of course that needs to happen. And if I had the hundred richest people in the room, I would say, like, you know, you should do this, otherwise, you know, it's going to be French Revolution on your ass. And you you know, you don't want that. So I think everyone gains from a more equal society. Yeah, and that's a this the discussion needs to be had basically.

Anders Arpteg 2:16:32

What is it? Uh Lex Freeman always says, Um, power corrupts, but absolute power corrupts absolutely. Exactly. If if some person or company gets absolute power, it will corrupt them somehow.

Abundance Versus Dystopia Endgames

SPEAKER_00 2:16:45

And it also destroys innovation, right? I mean, if they don't have anything to compete with, then they probably just okay.

Anders Arpteg 2:16:52

Coming to this last question, then uh at some point we probably have AGI perhaps very fast, and uh perhaps a bit later, and perhaps technology is moving very fast and adoption a bit slower, but still at some point it will happen. We can imagine then two extremes here. Either it can be the horrible future, and we have uh the Terminators and the Matrix and machines trying to kill us all, which you know, given the geopolitical situation, is not unfeasible either. They train on us on our data. And then uh we can take the other extreme, and that's the uh utopian future, and we will have uh AI that solves the climate crisis and uh finds a cure for cancer and fix the energy problem and basically creates the world of abundance that makes price and of services and goods go to zero, and we don't need to have a capitalized uh society anymore. Wh where do you think we will end up? If we start with you, Tom, here uh if we do you think what's the chance of us moving in one of these two extremes?

SPEAKER_00 2:17:54

I mean they're they're um good arguments for both sides, I guess. I I heard this like um when we uh happen to step on an ant, we don't feel that much you know hurt, or we're not we don't care that much about the ant. Why is that? And science says that apparently it's because our intelligence gap is so big. So if you argue that we would create an AGI that is many times smarter than us, that AGI would probably not feel so bad if they need to remove a couple of people because we're too many and we're consuming too much and so on. So that's of course an uh a rather dystopian um scenario. Um but then you know I think we created in this liberal capitalism system that we have across all countries today. Uh, we have this system that kind of just drives innovation and there's no way to stop it. I mean, we can't go back and say, like, oh, we're gonna do planned economy now, and you know, even if that has its own ideas, you know, it's never gonna, you know, that's never gonna happen. So I think we just have to jump on a train. And I think being positive and like the abundance movement, for example, uh, I think that's probably the the way to go. And hopefully, you know, we create this super intelligent being that can solve our pro all our problems, and everyone can uh do what they like to do most, which is probably being with uh friends and family and enjoying your hobbies and brew beer, perhaps. Yeah, yeah, yeah. Lots of beers. More beer, definitely.

Henrik Göthberg 2:19:28

So we have beer Bretopia mission. Yeah, Breetopia, definitely. Darner, what do you think?

SPEAKER_07 2:19:35

Yeah, no, I I actually used to be in like the camp of um that, like, yeah, no, we need to like really put the brakes on this and like you know, embrace the climate movement and like draw down consumption and like do all of this really seriously, like otherwise we're gonna fuck up the earth. But now I think the irony is that like I think people don't like that Elon Musk is on his way to space and he's gonna spend so much money on spaceships, etc. But I think that he might actually save the earth's climate because of it. Because if you move all of this fantastic AI activity out into space because it's more efficient, you also don't have to like paperclip the whole earth. It can stay as a like little museum of the origins, and and like AI can go out and do stuff in space because it's just more efficient, better scale, and and uh yeah, way more inspiring to the AI.

Anders Arpteg 2:20:29

We make AI multiplanetary, isn't it? Exactly.

Henrik Göthberg 2:20:35

That's just giving us the biggest scam.

SPEAKER_00 2:20:37

Look how much it's over there.

Henrik Göthberg 2:20:39

We use ads.

SPEAKER_00 2:20:40

You want to go in that direction? It depends, you know, it's singularity can happen as well as we can follow on a hard drive or something. I haven't thought about that.

Anders Arpteg 2:20:46

That's a really profound, I think. You know, we actually create an outlet for the AI in some way, yeah.

SPEAKER_08 2:20:51

And then like the whole market economy or like the consumption part that goes with it.

Henrik Göthberg 2:20:56

And we just get 1% of the profits from the AI that's written into the something like that.

SPEAKER_07 2:21:06

I mean, he has even argued that this Dyson form that he plans to send out, like people say that jokingly, I think, because he's really planning not to circle it around the sun but have it somewhere close to Earth, primarily. And you can even use that to shadow some of the sun rays coming in and thereby dampen the whole like greenhouse effect, or the effects of the greenhouse effect, if you will. So I think there might actually be like a climate solution in the the whole like in the background here.

Henrik Göthberg 2:21:35

You actually solved the climate. We went to the abundance, we solved the climate crisis in a completely different way.

SPEAKER_07 2:21:41

Yeah, because if we build out AI like as much as we will want to in the next 50 years only on Earth, I mean then we're leaving. Then we'll see it.

SPEAKER_08 2:21:49

Then we fucked it up.

SPEAKER_07 2:21:50

You can forget about taking a stroll in nature after that. It's just gonna be like pure industrial area, pretty much. But but yeah, going to space is a great solution, I think.

SPEAKER_04 2:22:00

Interesting.

Henrik Göthberg 2:22:01

Very inspiring.

SPEAKER_04 2:22:03

Yeah, I like that.

Anders Arpteg 2:22:04

Oh, that's a new angle. We tried this question like a hundred times or something, and uh yeah. We'll be we're building data here. Yeah. Going to train an AI on this.

Henrik Göthberg 2:22:17

We almost have the answer now.

SPEAKER_07 2:22:18

I mean, this is already streamed to the internet, right? So we're already contributing to the data. Contributing.

Anders Arpteg 2:22:24

We are we are it's being rocked. Okay, thank you so much, Banner, Sataval and Tom Strom, for coming here. Uh, loved uh that you could come here and speak about so many amazing things. And I hope you can stay on and continue to speak after uh after work uh and the camera turns off. Uh so thank you so much for coming here. Thank you so much. Thank you for having us.

SPEAKER_07 2:22:45

Yeah, thank you. The pleasure was really ours. Uh it was great to be here. Uh nice atmosphere, good drinks, and good company. So thank you.