84. What do DeepMind & Demis Hassabis teach us about the new era of Category leadership?

0:00Welcome to The Difference Engine, the show for tech founders, investors and innovators.

0:11Today we'll be analysing the anatomy of hype and how it can be used for category creation. Should you chase the noise or just ignore it? But first, what does the Google reshuffle teach us about a new way to create category leadership?

0:29Today we're looking at a story that, on the surface, is about an executive reshuffle at Google DeepMind. But I think there's much, much bigger questions beneath it. Are we seeing the emergence of a completely new model for achieving category leadership in technology? Because historically, big tech's instinct has been pretty simple. Find the world's best talent, acquire them if necessary, give them enormous resources, and then try very, very hard to keep them inside the organization. But what happens when the reality is that you simply can't keep them forever? Right. So that brings us neatly on to Demis Hazabis, a UK entrepreneur, some would say category definer, certainly a major scientist in the world of AI, who's not really leaving Google DeepMind.

1:14What they say is he's stepping away from the day-to-day CEO position and moving into a chairmanship role, but he's also keeping his chief scientist title at Alphabet and as beginning to build out isomorphic labs, which is the AI of the real world as it's fashioned. So rather than saying, I'm out, we've done everything we need, I'm going to be a CEO somewhere else, he's actually saying something slightly different. And Google's sort of said to him, I guess, what's your next challenge, Dennis? We'd like to help you with that. And that's a very, very different philosophy to what we've seen in the world of software heretofore.

1:50Yeah. And it's about talent management, but it's also about category creation, Because I guess, if you think about it, DeepMind itself was essentially the original example of this model. You know, Hassidist founded DeepMind, then built it into one of the world's most important AI research organizations, and then Google acquired it. Now, the traditional acquisition playbook would be bring the founder in, integrate the company, and eventually make that founder part of the corporate machine. Now, I've never really believed in that because people are entrepreneurs by nature and they never do particularly well being a large corporate servant.

2:27So at least now we're seeing something different. And it's the realization that, you know, when you've got a real, you know, and let's face it, let's call Hassabis what he is. He's a genius. Running the machine is really not the best use of his time. Well, that's really, I guess, the problem here. It's like when you've got global talent like Hasevis and one wonders what would anyone do to keep Elon Musk on board, another genius. But the question wasn't how do we retain Demis' services, traditional employee-employer relationship. It's more like how do we keep Demis doing something that's important to us and to him strategically?

3:12And those are two slightly different questions. Indeed they are. And I think this is where the category leadership argument gets really interesting. Because if you're Google and the size you are, the influence you exert, the money that you make, you don't necessarily need to have every great scientist spending the next 20 years climbing the corporate hierarchy. what you need to remain in place is is is to be the place where the world's best people can take on the world's hardest challenges i mean that that's ultimately what you're there for um and that should be extended to the idea that even if that means that their relationship with you the company actually changes quite considerably yeah and um you know i can't help of thinking about sports analogies and how folks at the top of the game, you know, your Messies, your Rinaldos in the soccer world, for instance, how people bend and change the rules to give them a different cut of their fees, to work collaboratively with them in terms of setting up other clubs.

4:20Another example here is Jeff Dean. So Jeff Dean, a lot less known than Demis, has been at Google for three decades. He's paid his dues. And now he's leaving with a group for very senior executives to do something called discovery loop. But guess what? Google's investing in the company. That's fascinating because in the old model, they have to part ways. People will be saying, is he pushed? Did he jump? Now we've got something more subtle. And the question becomes, will Google be the right place and the right partner to help Jeff Dean build his next thing? The basic philosophy is a profound shift.

4:54If you think through our decades in tech, what tended to happen to people that were acquired and we've seen some real big bust-ups here is that you know if you decide you're going to leave even after you've completed your own outs or you know whatever strictures you were under when you joined the company or even as a long-term employee like dean um okay so if you leave and we think you're going to take some intellectual capital with you uh we're going to sue you and we're going to slow down your company and destroy it if we can now it seems that the at least at Google, as an emerging mindset here, which is, well, if you're going to leave, let's make sure you leave to do something ambitious.

5:40And let's find a way for us to participate and, of course, benefit in the long term. Yeah. And to your point about lawsuits, there's one flying at the moment between Apple and OpenAI about former employees and taking IP. That's just nastiness. It's a distraction from cracking on. And in this world of AI with new inventions new models and new takes and almost you know trillions of dollars changing hands almost weekly as announcements are made this is a different situation than we've ever seen before um yeah so from a google perspective you get you know you get the capital you get the infrastructure you get the compute you get the talent and you get the networks and from the employees point of view that's also a good outcome and just in case anyone wants to see what a less good outcome looks at you can see the recent um you can see the recent piece that's going on with bending spoons and air table and you've got you know disenfranchised frankly pissed off former employees saying this was a terrible outcome now that's not what the founder thought but that's what they thought they didn't understand that they were just salaried employees but at this top level that you're talking about we're talking about here the real category creators there's a different thing going on something perhaps we haven't seen before with both sides seeing up some upside i guess yes the question is how does this new kind of close spin-off architecture look we've not seen it before and will it work we sort of don't know but it would appear to be a way that innovation can be accelerated faster than than if somebody was nailed into an organization but i do think there is a kind of counter argument and of course this this this plays to um both paul and and i's past where we all think about what's the actual narrative behind this.

7:24You know, it's been very cleverly spun, this certainly, but, you know, could it actually just be a rather clever way of putting a positive spin on a talent exodus? Because Google is also losing some extraordinarily important AI people, guys like Noam Shazir and John Jumper and others. And Google is, of course, operating an environment where OpenAI and Anthropic are moving incredibly quickly. So you could look at this and say, this isn't a new model to deliver category leadership. It's a company with actually an ever more weakened core trying to make the best of actually losing people. Right. So to develop the sort of, let's say, the bear case here for what's going on, maybe it's true that there is a talent retention problem.

8:19I find it ironic because one of the accusations, largely at Google, but at other places too, and you see it in the sitcom Silicon Valley, nicely portrayed, is that these guys were going out acquahiring just to put people on the bench. They were not so much hiring talent that they really wanted as starving their competition of talent that they didn't want to be competing with them. And maybe here is something analogous, because if they lock up these people into a loosely affiliated organization structure, they might be denying competition from happening that would be negative to them. So, you know, look at Hasibus.

9:01His comparative advantage is that he thinks a lot about AGI. Now, whether or not you believe in that is very contentious. But having him spend every day managing an enormous organization that you own maybe isn't as valuable as having him work on something and if that works out hey there's upside and if it doesn't work out at least he's not competing against you yeah but you know at least he's possibly doing what makes him happy i suppose and i think that's a an important lesson for category leaders generally um it's you know and i've been through this a number of times is that it's a mistake to believe that the person who is great at creating the category should necessarily run the organization forever and actually it requires different skills to do that um you know the person that has the mindset that can can see the future and can see the opportunity and then go out and create it um isn't necessarily the person who should run the operating machine that ends up delivering it right and and you know so what happens to those who remain and i'm going to murder this name but perhaps that's where um um karai kavukoglu i hope my turgish is okay there um you know the operational leadership moving to someone else who's been deeply embedded this is important uh because if the message is all the talents exited then you've got a real issue but if some of the talent stays behind then you've got a shot um and somebody else running the engine can also uh often reinvigorate that engine so in many ways that the hr move here is super clever it's sort of you know it's the word it sort of sounds obvious when you say it like that but it you know i think if we look back over the history it's really quite radical in founder-led technology companies because because we tend to equate in tech uh leadership with with control and the ceo of course is and should be the person with the most control but perhaps the next generation of category leaders will be will be less interested in control and more interested in optionality.

11:07Wow, I love that. But that is quite a change. I mean, if you think about your classic tech leaders, Ellison and Jobs, they were all about control. And the technology companies that they invented did look quite like a pyramid in many cases. If now things start to look less like pyramids to be climbed and more like networks to be maximized, that's a brand new model. And you always want the best people connected and you lose them when they leave your ecosystem. system. But the point I think you're trying to make is your talent doesn't need to necessarily report to you in that model. They don't necessarily need to be on the payroll.

11:42They can still be regarded as founders, investors, and importantly in this AI world as AI researchers, who knew that word would be so valuable. You can talk about spin outs, they can be your partners, they can be suppliers, they can even be your frenemies if you stay on the right side of the law. And so the strategic asset there almost becomes that network itself. Yeah and why is that particularly important right now? Because guess what? AI accelerates that. The half-life of any organization is inevitably going to get shorter and shorter and shorter and shorter and the reality can be maybe you can spend five years maybe building a world-class AI team, and then suddenly the frontier you're aiming for and frankly your ambition moves somewhere else.

12:35And so you may get an AI researcher who was perfect for developing foundation models suddenly might go, actually I'd like to work on the application of AI to biology. Or another one suddenly decides they become really fascinated by software engineering they want to automate that or another one has seen what the ceo has done and decides that they want to build a company but you know if the parent company says well no no no no you're staying here doing exactly what you want to do um because we need you to do it sorry they're just going to walk the talent is going to walk and yes that's down to like the footballer analogy and the musician analogy which we've seen seen over the years yes i suppose in contrast that if the response is how can we help what's the biggest problem you want to solve then that's just a nicer conversation and you might actually get more out of the relationship not necessarily the employee but the relationship and when you think about the logic of stuff that's not possibly worked as well as as it should have done like carve outs spin outs you know corporate venturing all of those things where they're trying to keep the best of both worlds this seems to be like just a more thought through human centric version of that um so yeah i'm fascinated to think of this distinction between what you can get out of an employee and what you can get out of future relationships if you strike a deal like this what that means i really feel this is going to become incredibly important and i think the whole history of the it industry has been leading up to this And I think it can help us define what it is, what it takes to create category leadership.

14:19Weirdly, not actually obsession with owning the category, but being the organization that continuously creates the conditions for new categories to emerge. You're constantly reinvigorating talent. Yeah, yeah, yeah, yeah. So go create something new, backed with the money that you'd have to go out and raise anyway, which was sort of where corporate venturing, I think, went a bit wrong, because it was a bit safer than that. You know, these guys, Demis et al. have got reputations to uphold, and you know damn well, they're going to give it everything they've got. I suppose that's a real test for Google.

14:54Can it keep all its, you know, some would say boring, some would say failing, some would say it's yet to shine, Gemini core AI business moving while it's allowing all of its ambitious researchers to go plow their own furrow? Yeah, there is a danger in that, isn't it? If you turn every departure into some form of spin-out, you do end up hollowing out the core. Of course you do, unless you're constantly bringing in new people. at some point you need to go hang on a minute there are people we need to keep inside the organization who are actually building the products and in the mothership you know you you can't suddenly decide that you're going to turn into some form of venture capital arm for former employees yeah but i should imagine there's a there's a i know there's a there's a queue around the door to join most of these ai companies and even even quote unquote google i'm sure there's a you know i should imagine being a recruiter there is like shooting fish in a barrel as they say but category leadership requires more than just talent you've got to have a bit of infrastructure you've got to have amazing products it helps and they say distribution always wins it helps if you've got the distribution and a commercial engine meaning not just a founder but someone who can go out and sell that vision i'm beginning to wonder if if what we're you know what we're thinking about here is that we're moving towards a sort of a two-speed model so So you've got different things which matter inside the company and around the company.

16:21So sort of inside the company, you've got scale, products, infrastructure, and of course, good old execution. And outside the company, you've got the research, the spinning out, the founders, the moonshots, and entirely new categories. Yeah, I've got to admit, from a European perspective, a little bit of nervousness about the duopoly that's forming, keeping Google out of it for a minute. but big tech's got all the advantages they're enormous really enormous look at their balance sheets they've bought the compute and the power and the distribution comes with their established place in the market clearly they are able to convert you know customers into ARR very efficiently and in many cases they've got decades of winning so that institutional knowledge that comes from which of the bets I want to make so you know why would they why would an ambitious researcher want to risk that when they've got all those advantages.

17:16And I worry about competition globally if this works and it freezes out genuine, innovative, disruptive companies from coming at these guys. Again, this could be a natural extension of capitalism, which is just a new way of concentrating power. So even if they go, the third option, which is really go build the next big thing and we'll back you. But that would form a real change in the psychology of talent management because it really does come down to stop worrying about stopping your stars from leaving. It's how do you make your ecosystem so attractive that exiting a position in a company doesn't mean leaving the company behind.

18:11Yeah, it sounds like winning it both ways. What we know from tech is that a new wave will come along, whether that's quantum computing or whatever. So this may be a short term solution to the talent drain problem that AI causes. What are the bigger lessons do you think we've learned from the departures of Hasibis and Dean for now? I think what we're seeing is, if you look at the individuals, one is moving further into the scientific future while remaining deeply connected in alphabet. Well, that's the spin. And the other is leaving to build something new, but with Google as an investor. And, of course, that is the spin.

18:50Now, think about it. They're pretty much two different versions of the same idea. That relationship can survive the employment contract or the traditional employment contract, which was a sort of master peasant relationship. It's much more equal. It's much more networked. And that could be the new model for attaining category leadership. Don't try and own every great person. Own the ecosystem in which great people can do their best work. Yeah, because in the next decade of technology, the category leader might not be the company with the biggest team. It will be the company with the strongest gravitational pull around talent, capital, technology and ideas.

19:46Building a technology company is brutal. We know that. It's years of early mornings, late nights, working weekends, long hours, lost sleep and neglected loved ones. You keep going because of a relentless belief in what you're creating. At Hambleton Partners, our senior advisors have walked that path, experiencing at first hand the challenges that technology leaders face every day. We know selling your tech business isn't just a transaction. It's one of the most important life decisions you'll ever make. That's why we take a different approach to technology M &A. We combine deep transaction expertise with hard-won industry savvy, helping founders navigate every stage of the transaction process with empathy, insight and the confidence born of deep experience.

20:34When the time is right to explore what's next, work with advisors who understand every step of the journey. Hamilton Partners.

20:47We'd just like to do a bit of tackling of one of the questions that technology has been arguing about for decades, well, certainly as long as I've been involved in technology marketing. Can the hype around a company or technology actually lead to category leadership, or does chasing hype actually distract you from building something much more enduring? Well, yeah, the word hype, I think, causes some confusion. And certainly, I'm looking at a lot of hype-y words, let's just say, that are out there right now and things that come to mind just off the top are, you know, rogue AI, that's a big one, token maxing, tokenomics, AI control planes, industry 6.0 guardrails.

21:29I mean, you can tell this tech sector is just rich with like the little buds of new categories. Yeah, but I think the thing is, when are those buds important and how do you identify them? That's all about that, yeah. We came across recently a really interesting little bit of research. done this summer by top-ranked VC Sequoia. And what they did was that they chose to start digging into about 20 years' worth of hype data. The way they defined this hype data was, you know, what were the projects that generated the most attention in hacker news, right? And they went right back to 2007, which, of course, you know, is pretty much two decades' worth of evidence.

22:13Yeah, I love these COD science type research projects. If you think about some of the data points in the original Play Bigger book around that, there was a lot of subjectivity around which were the category defining companies. Nothing wrong with it. It's a valid research format. But what these guys have done is they put the most valuable company according to them in each year alongside the most hype topics. So let's get going. You know, that should let us compare a little bit to what Hacker News thought was hype versus what actually happened. What have you got for us? One of the things that jumps out almost immediately when you look at this inverted commas data is how rarely the idea of category leadership and hype actually tie up.

23:01So if you think about Airbnb, you know, founded in 2008, what was the top hype topic at the time? anything to do with Google. Uber arrived in 2009, while the conversation was all about low-level programming. And Anthropic, founded in 2021, and all everybody was talking about was crypto. Yeah, so pretty powerful counter-argument to the idea that the smartest founder looks at the hottest topics and builds there. You've seen cloud washing, we've seen green washing, we've seen AI washing. If you define chasing hype as building what everyone is talking about right now historical evidence says that ain't great hang on let me push back on this this this for a moment um perhaps we're looking at this through a survivor bias lens um you know maybe it was that the companies that successfully chased hype aren't obvious from this particular comparison or maybe even the the relationship between hype and the company creation is much, much more indirect.

24:08I mean, that's an important caveat, and I hate myself for doing this, but I'm sort of going to query the data here. The data doesn't really prove that hype has no causal relationship with company creation. It shows the most valuable company formed in a given year is rarely a direct response to that year's most popular topic. But we're not saying that hype is useless. were saying that build whatever is number one on the internet chatter today is probably poor company building strategy. Yeah, you don't necessarily need to chase today's hype. And of course, we would never recommend that in any category discussion.

24:45We're looking for the thing that's just over the horizon, the unmet need. It's a key part of how you build great categories. And you definitely, though, don't want to ignore hype signals. They can help you come to the conclusion of what's next. Yeah, because if you think about it, new trends often announce themselves years before they become mainstream. We love our history here, but on this one, if you look at LLMs, they first cracked the top 15 of discussion topics back in 2016, and they didn't reach number one until 2022. Yeah, it feels a bit like Tom of the Pops. That's a five or six year lead time, time uh which is remarkable AI coding uh you know my friends were doing that back at university several decades ago but it entered at number 12 in 2021 and now it's top of the list in 2026 its time has come crypto appeared in 2011 before its major attention peaks with NFT etc 2017 and 2021 I would argue it's better its best days are ahead of it oh that's uh that's really really interesting so I guess what can we get from this at this stage is that you know today's obvious trends at some point with like yesterday's niche conversation which sort of suggests to us I think that internet sub-communities if you know where to find them and follow them will act as an early warning system for technological change and what is going to go on in category creation in future Yeah, and I'm sure that all the hedge funds and VCs out there have got all sorts of systems to detect that now that are AI driven.

26:25Let's go back to crypto for a sec. So crypto is a great example because the underlying idea of, you know, blockchains and systems records, systems of proof of work, all of that great stuff were generated and actually almost perfected years before, you know, Coinbase and Bitcoin and all of that stuff became known in the public psyche. So being early to a trend isn't necessarily the same thing as getting your timing right, basically. So it makes me think, you know, perhaps the right question isn't, you know, what's the hottest topic? It's really what's moving. And it's that vector which is most interesting.

27:07So, you know, if something goes from nowhere into the top 50 and then the top 20 and then the top 10 over several years, that progression may be a lot more informative than simply looking at what's number one today. Right. Don't study just the peak, study the slope. Could be a much more interesting signal than something that suddenly became number one just three months ago. Yeah, yeah. But that, I think, in itself brings us to possibly a third observation, which I think is particularly fascinating. Long-term hype actually appears to be a durable signal. right just stay with me on this so what we laughably call the muskverse um has been a top 15 topic on hacker news for every one of the last 14 years all right that's an elong time so i did it oh yes i did i did actually sadly see exactly what you did that so sustained attention is unusual.

28:07I would add another category, which is misunderstood hype. The crowd gets it directionally right, and they get excited about a technology, perhaps too excited, but something goes slightly wrong in terms of timing. The application doesn't fulfill its expectations, or the business model gets disrupted by something new. For instance, you know freemium sass for for example yeah i i do think that's where founders can get themselves into trouble you know you see you know this massive technological trend and assume that the obvious company attached that trend will automatically become the category leader that may not be the case yeah so category leadership requires a bunch of stuff uh you know to sustain it basically product quality distribution customer adoption uh brand capital um capital itself ecosystems that you build up moats around switching costs as well as the ability to define and probably the most importantly to define the category in the customer's mind which is sort of hype right yeah yeah we're sort of bringing ourselves around in a circle i mean you know hype hype clearly clearly can help create the conditions for category leadership but that won't necessarily determine who wins you know of course if you've got hype around your company you're going to be able to attract talent investors customers you know developers press um and you know that's fine that's lots of attention but but somebody still has to get practical you've got to convert all that attention into into some form of actually durable competitive advantage you know a story is not a strategy in that sense yeah it's so hypes you know it's not a moat that's what we're saying here um but the moment a ironically the moment a potential category becomes obviously attractive uh the competition increases and so does quite clearly the hype the two are linked yeah there's an interesting paradox here i think you know hype can make a market more valuable while simultaneously making it harder for any individual company to win so once everybody sees the opportunity suddenly you've got suddenly dozens of startups incumbents investors consultants and conferences focused on the same category you know blockchain a few years ago that was exactly what was happening yeah and so i think we're saying don't chase the peak because it's often too late when you're chasing it study the trajectory um and look for those technologies and particular customer behaviors which are changing or about to change because there's going to be in there a compelling customer problem.

30:47And if you find one that's unmet, that is your gold. That's your North Star. Yeah, right. So, yeah, in other words, you know, the sweet spot is somewhere between nobody cares and everybody cares. Hard thing to figure out, right? Yeah, it's uncomfortable because you're making decisions before you have any form of certainty. But, you know, obviously, that's also where the biggest opportunities are going to exist. And just to, like, you know, maybe join the dots here a little bit. I mean, Hacker News, again, I don't want to query the data because the data is the data, but is Hacker News the right place to be looking for this?

31:20You know, engineers think a certain way. Category creators think another way. But maybe, you know, these Internet sub-communities can sometimes see things, certainly technological things, early. Whether they can weight them correctly, I have my doubts. Yeah, yeah. And I mean, you know, other sources are available, such as, you know, X and LinkedIn. And, you know, Hacking News might be particularly good at servicing technical trends. X, I guess, on the other hand, might capture, you know, emerging probably quite sweary cultural conversations. And LinkedIn, of course, will tell us something different about enterprise adoption.

31:59Oh, God, you know, not to mention self-interest and so on. Yeah, and the LinkedIn AI slot button, I mean, that's been created for a reason, right? There's just so much nonsense on there. And so it's easy to create fake hype, fake interest with AI slot. So this whole hype topic is a tricky one. And creating a hype score, I think, is, let's just say to the Sequoia guys, ambitious. You know, communities tend to identify stuff early. Whether they act on it is completely different. So let's bring this back to our original question then maybe. Yeah, so I mean, can hyper-and-a-company lead to category leadership?

32:41I think my answer would be yes. But I guess as any scientist will tell you, the causal relationship is much more complicated than people usually assume. It's an important distinction. Attention is obviously not just important, but it's an extremely effective distribution strategy and it's extremely effective or has been heretofore for investment. attention um so so you often get the case where a you know a company is riding a hype wave and it's teaching the market what a category is creates a language defines the problems establish the benchmarks and becomes a reference point but misses out for some reason on category leadership and therefore the majority of the economics um so it's a tricky tricky one to figure out yeah there's all that danger of becoming too focused on attention and you're optimizing actually for being talked about rather than actually being useful.

33:36You can end up with a very valuable media company and a very mediocre technology company. Yeah, I've seen that so many times, both of us. Let's move on to the investor side of things. That's your favorite subject. What do you think, you know, if you're an investor looking at a highly hyped company and lots of people listening to the pod have got companies they want investors to look at, what are the three questions you think that um that you need to ask yourself oh right um so first thing is is all about reality it's is the underlying technological um market trend real um is the addressable market expanding um and even if the market becomes enormous why does this particular company have a credible path to winning it get really specific there yeah right so let's let's take them in order so huge market clearly doesn't necessarily produce a huge company um something that the ai companies that are spending all their money building data centers might want to remember um then you know the companies that that um create the most um with the largest TAM, they may not capture all of that value, right?

34:48So early category leadership and enduring category leadership, two different things. And so almost the timing of the hype is really important. The company that gets the most attention in the doors isn't necessarily the company that ends up dominating. And you've got your examples here of Mosaic, when Google took over, and there are many other browser companies, AltaVista, Lycos, et cetera, we could talk about from that example. um yeah so hopefully that deals with the um the time question um yeah why why is it that the companies founded in recent years are more interesting we don't have the benefit of hindsight yet but why are we attracted to those guys is it is it hype what's what's going on there we don't really know which of today's visible companies will become the enduring leader no we don't what What we do know is AI coding is the number one hype data winner in 2026.

35:42You're talking cursor, you're talking lovable, you're talking codex, you're talking a lot of things, yeah. I mean, classically hypey. Five years from now, we may look back and discover that the biggest company from this wave was the company that everybody was talking about today. or more likely, I think, from the evidence, will discover it was a company that has barely registered on the radar. That's an amazingly powerful and encouraging thing for folks who are chasing a category and are worried because of hype that they don't have a shot. So I guess your message is don't build for the trend that's loudest today.