Reid riffs on acquihires, agents, and new startup norms

0:28I'm Reid Hoffman. This week, in part one, we're sharing my live conversation with Reid. And next week, we'll dig into those listener questions. So without further ado, here is Reid Riffs Live.

0:43So I think we're going to start off with the hardest question. One of the things Reid always says is anytime someone says, oh, here's what's going to happen in 10 years, they're probably an idiot. It's really hard to predict things that are 10 years in the future. But I feel like all of the time on Twitter, people are always saying, don't build for the AI we have today. Build for the AI that we have tomorrow. This is moving so quickly. You need to anticipate where AI is going to be, and you need to build for that. But how do we correctly anticipate what tomorrow is? Is it the AI of next week, the AI of next month, a year?

1:19And I know this is an impossible question because it matters the sector. Is it SaaS? Is it consumer? but just give us a little idea what AI should we be building for? And two hours later, we finish the answer to question number one. Oh, yeah, read 30-second answers, please. And so, look, as the question well sets out, what you have to look at is the classic kind of startup things, which is what is the zone in which your competitors are operating in? What is the zone in which the speed in which technology, AI in this case, is changing for this? What is your go-to-market strategy? There's a set of those things.

1:53And so the answer will be different, not just on sectors, but it's also kind of a question of, you know, making some predictions about which of the different threads both we see currently that are iterating and improving and also which ones we see coming in. And then at what speed. So, for example, we all know that there is an intense amount of effort around everything that gets to do with large language models, you know, coding, text, work, productivity, et cetera. and part of what that means is you go, I know what it's going to be in three years. It's like, no, you probably don't. And so you really have to be dynamic and really updating quickly for that.

2:32If you're in multimodal and you say, well, for example, one of the things that I learned from Ethan Mollick that I thought was super interesting and many things, but as an example of where Ethan was showing that people are just not even, like are way under utilizing these tools that exist today, he said, hey, you know, like one of the difficult challenges is construction projects. And like, namely, is it on track? What might be running slow? Et cetera, et cetera. And he literally uploaded the construction plan to an AI, put, I think it was like 20 cameras around it, and then asked, how's it going?

3:07Like, had it every day, every hour, ask how it's going. And it gave a pretty accurate construction report of this is what's running slowly, this is how the work really ran today, et cetera, et cetera. And that already exists in multimodal. But multimodal is probably one of the ones where, you know, where it's because, you know, error rates and all the rest. And where does that matter? And, you know, like a monitoring thing, you know, you could be wrong and you can go check it and so forth. So it's one of the things that multimodal is in the middle. And then you've got things that are a little bit longer out, which is even though people are intensely doing science and everything else, it might be the where will the various forms of science acceleration models happen.

3:46And so you get all of that factor. but a lot of it's the kind of classic, you know, what's the speed of the competition? Like a lot of entrepreneurship is about competition, not usually the mistake of thinking it's per competition with the large players. Usually it's other startups and, you know, how many of them are they on the same path you are? What is the go-to-market? What is the defensibility of the thing that you establish once you're there? And all of that plays into it and the timing on the AI model and trajectory of where the hockey puck is moving to depends on that. Can I just pick up on one thing?

4:21We have a lot of founders in the room. You said the mistake that some people make is they think the startups are competing with incumbents, but they're really competing with other startups. Say a bit more on that. Well, so mostly all organizations, small, large, have, you know, call it a small number of priorities. You know, the memory thing is most people can remember seven plus or minus two things with a little bit of an emphasis on minus two. And organization tends to be like three things, plus or minus two things. And so, for example, if you're trying to do is say, hey, I'm going to go build a new, call it office product suite, and I'm going to do it that way, and I'm going to go to market, you're going to have a challenge because the giants are actually, in fact, really fighting with you on this.

5:08It's like I'm going to do desktop search. Now, by the way, of course, like one of the things, If you haven't started experimenting with putting the chat GBD plug in to interrupt your search and do it, you should. You learn interesting things from it in terms of where it's ready, where it's not, and all the rest. But the typical thing is, well, does Google, Microsoft, Amazon, Apple have a small group working on your startup project? Yes, they probably do. And who cares? Depending on if it's not one of the three plus or minus two things that the organization is doing. And so, as a matter of fact, it's almost like a little bit of validation if you say, hey, I'm working on this thing and there's a group at HyperscalerX that's doing it.

5:49Now, the asterisk, the reason why, that's the general software startup advice. My asterisk is a little bit of, there's also, there's an intensity to the play here in how AI software is constructed that involves scale compute, scale data, scale teams. And it doesn't mean that startups can't do their own models and can't be doing things. But if you're suddenly in the, I have to be playing scale across those vectors, then all of a sudden you better have a good strategy of the game there. So let's talk about those hyperscalers. Let's zoom out. In recent weeks, it's been a little nuts. We had Meta's investment in scale AI at 49%.

6:30We had OpenAI pay$6 billion for Johnny Ives startup. There's talk of Nat Friedman and Daniel Gross, Aquahire, to go sort of be the meta dream team. Is this like a new era in the AI world where in the past we had acquisitions and now we care more about talent than we do about product? Is this a new wave or is it just a blip that we're seeing? Well, it's a new wave partially occasioned by what I have, in a couple of different ways, said is an overly aggressive FTC, where the notion is we're trying to prevent future issues that look like monopolies versus current ones. And so that kind of thing is actually very anti-business, very anti-venture, very anti-constrained.

7:25And that required a bunch of innovation for some new deal types. So it's new deal types have now, I think, been definitively added. But not having been anywhere in the room and having any specific intelligence. But I suspect that if Meta thought that they could have just bought scale, they would have just bought scale. It's much simpler, doesn't require all this, other sorts of things. Now, there are a couple of footnotes in these deals that are interesting. The basic thought is we're in the largest technological revolution that we've had in human history, that the outcomes will create companies that are in the tens, hundreds, billions, and even, now we say trillions of dollars in value.

8:06And so key talent might actually, in fact, be worth not just millions, but tens of millions, hundreds of millions, and maybe even billions in order to do that. And that still fits within the envelope of how, you know, talent is generally valued within companies. Because the framework is usually like, you know, if I pay you, you know,$10 million because I'm expecting to make$100 million off the platform in terms of doing it. And that's part of how shareholder value, collective value in the company, and all the rest of the stuff is built. But even with that, because of the anticipation of value is so high within some of these AI companies, that that's the bets that are being made.

8:47And it's not crazy. Should the people in this room do anything different? Like, obviously, there's still the path to IPO and the IPO market is opening up a bit. But do you build any differently when you're expecting this path versus a pure acquisition? You know, I was actually just before this talking to, you know, one of my partners who works here in New York, Seth Rosenberg at Greylock. And I was just talking to him about this. And part of the thing is to say you always build for an eternal company. You build for it's going to go public, it's going to be big, it's going to change an industry.

9:21And you change other things based on it. Now, the pattern of how you're building might change. Because the pattern of how you're building might be like, okay, for example, one of the things that's a near certainty is within a small number of years, could be as small as two, could be as great as five, that if you're a professional and you're deploying and you're not deploying with multiple agents in what you're doing, you're under-tooled. It's a little bit like saying, I'm a graphic designer and I don't use Figma or Photoshop. Where it's like, no, not really. You're not a graphic designer. You might be something else, but you're not a graphic designer.

10:03Or I'm a professional. I have neither a computer nor a smartphone. It's like, no, no, that doesn't really work that way. And so that will be intensely there. And that will change certain patterns. For example, one of the things that we've got classically within management is this notion of there's individual contributors and there's manager plus doers and there's managers of doers and then there's executives of managers and so forth. And when you go all the way down to the individual contributors in this universe, here's an interesting question. I won't make a prediction as per your earlier thing.

10:41But when will MBAs start, business schools start teaching classes on how you manage agents? Right? Because each individual contributor will have to have managing agents as part of their skill set. I thought you were going to say, when is the MBA going to become irrelevant? And perhaps the answer was 10 years ago. Well, I usually say that my degree in philosophy is more relevant than an MBA. Exactly. Exactly. Exactly. The philosopher entrepreneur. Okay, so going to go back to, all right, if we're building today, we need to think about the AI of tomorrow. Everyone's talking about what your moats are.

11:19It's like it used to be, was it network? Was it product? Today, it's like the magic of AI. You think something's great, and then next week, it's just table stakes. So in this new era, is it a proprietary feedback loop, your distribution wedge? What is the moat that these companies should be looking for? Well, so first is most of the classic modes still apply. You know, you get integrated into an enterprise workflow, and part of what happens, that's a defense. You have a network effect of, you know, of multiple sorts. And most people, of course, think network effects. They think LinkedIn and, you know, Facebook.

11:55Never heard of it. Yeah, whatever. This is a small company. And so, you know, they kind of think all those things. But there's also various forms. For example, one of the very original network effects in modern software was in Microsoft's Office product suite between the data file formats and the application. So there's lots of different places. And one of the, of course, key things to look at is where might there be network effects because they're one of the really key things that happen. Where might there be network effects that are new patterns within the AI thing? And most of the pitches that I've seen on that so far are more kind of impressionistic.

12:32you know, Jackson Pollock-ish than systematic, but, you know, I think there will be some. Obviously, you know, some part of the question is, you know, as you're solving certain kinds of problems, certain kinds of data set that really align to that will be very relevant because, you know, surprisingly, large models, GBD4, Gemini, Copilot, et cetera, can actually do a fair amount of medical stuff because there's just so much content out there on internet but as you begin to get to more kind of more specific things like okay drug therapy you know and and and reactions the data sets on those things actually in fact really matter so there'll be various things on data sets and so there will be some new patterns and obviously one of the patterns that are that the hyperscalers in particular focused on is scale compute you know we're still not quite sure where the scale compute really starts delivering sharply asymptotic value it's it's not to say that there won't be and all the rest and that's actually one of the questions that get into all the discussions of this and and by the way sometimes you know as you get to different orders a level of magnitude getting the models to um to to you know kind of uh congeal can come together in the right way actually has some added difficulty.

13:54It's not just apply 300 ,000 more H100s and everything is great. There's ways to try to make that happen effectively. But that's another one, and that's actually one of the things that makes the software game a little different because you say, well, actually, in fact, I've got a self-learning system that creates the real competitive advantage, but you actually have to have X hundred thousand H100s and when to do that, then that's less of a startup game and more of a hyperscaler game when you include OpenAI and Anthropic and other folks in the hyperscalers. So speaking of LinkedIn, which does have six times the revenue of Twitter, not that anyone asked, but just for the record annually, you launched LinkedIn over 20 years ago.

14:40And you had your co-founders, you had your first few hires. How does that look different than how AI teams should look today? Is it just less people? Is it mostly engineers? Do you wait a long time for that first business hire because you can do so much with ChatGPT? How is that different from 20 years ago? If you're starting a company from scratch and everyone in the company is not using AI aggressively, I mean, anyone who's not using AI aggressively, and I call it a seed or a series A company, I think you probably want to get rid of them. And if you're the founder, then get rid of yourself. So it's like, because it's where the puck is moving to and all the rest and understanding.

15:23It doesn't mean it'll be perfect for everything. It doesn't mean it should be used for everything, but that kind of cycle. And that means you'll have an amplification of productivity across the entire set. And so it allows a few new corner cases, like, for example, like maybe the cost of, as opposed to, oh, our minimum viable thing, like roughly speaking, I started LinkedIn, kind of call it the minimum viable company to really get stuff going was like kind of in the 10 to 15 people. Like, you know, when we at Greylock led the round on Instagram, it was 12 employees, right, as an instance. And so, you know, like that's kind of the set.

16:02And maybe now six people or five people or three people, you know, as kind of, you know, how that plays out. Maybe what you can do with one person is now much more amazing. And so there's some weird corner cases. But on the other hand, this is relatively small. And so, by the way, people who can't otherwise get seed capital or national can start early, it'll mean things for universities and university projects, you know, doing things. And so all of that opens up and it'll be very good. On the other hand, once you get into the kind of venture line, you know, the difference between funding 15 people and seven people is not that big of a difference or 15 and 25 people is not that big of a difference.

16:39And so I think it'll be less like, oh, we're not just now doing a lot more with the people. And I think that will be a plan. And so I think it'll probably be, you know, like getting to the blitz scaling, you know, kind of phenomena. Like the speed of motion is now, in many cases, especially when you have small teams, going to be absolutely important. The question of, you know, will people be fast following your particular go-to-market motion? You know, that's going to be relevant. So there's going to be a whole stack of kind of new patterns here. And by the way, in this regard, while there's new patterns, it's not like, oh, AI is the only time new patterns have happened.

17:14I mean, part of the thing that between my first startup, SocialNet, and PayPal, and then LinkedIn, is kind of like, how much do you have to have sun equipment, right? And Unix boxes that you're doing on-prem is just going down to the floor. Like, probably a lot of people here are like, what's a sun box? Anyway, but, you know, like, so it's a new set of tools, a new set of play, but that's not unheard of to change the game, the software game. Sounds like you're saying we're still going to need these, or perhaps not need, but once we get into the venture space, yeah, you're still going to have 15 employees.

17:48You'll start at six. You'll get to 15. You're just going to move more quickly. And so maybe that means that you're going to get to revenue more quickly or more revenue more quickly. So from the venture side, I mean, we've seen some eye-popping seed rounds lately, but forget sort of the Miramiratis and the one-offs. is venture just going to be investing earlier or are they going to be investing with more revenue and at the same stage? How does that change in the age of AI or the age of quicker work and more, just the speed is going up so fast? Well, I think there will be Crimea River for VCs, even though we're here with Village and sets in the audience and all the rest.

18:29It'll be harder because you can possibly blink and miss it more quickly. And so speed of decisioning, speed of making an offer, taking risks before there's super traction, et cetera. I mean, you tend to get this kind of barbell approach where either you have to go seed or you have to go growth. And the intermediate becomes kind of to some degree harder. Now, seed's always somewhat hard because you're like, okay, will they figure out kind of product market fit? Will they figure out something that will get to scale product market fit? And then in growth, a lot of it's kind of like, okay, well, how big does this get?

19:04And is it going to work? And then one of the things that frequently, once upon a time, Silicon Valley was very derided from and I think now is no longer, is you're not just taking product market fit risk or entrepreneur risk or competitive risk, but actually sometimes you take business model risk. And I suspect that one of the things we're going to see a lot of, kind of like almost like the parallel from the early internet to now with AI, is we'll see a lot of business model risk.

19:39so we're going to switch gears um my co-worker thanasi put this question in the question set because he knows what a huge fan of crypto i am so here we are we're going to talk crypto resurgence regulators uh have a have a changed posture towards uh crypto with the new administration We saw Circle IPO. It's up 620%, I think, as of today. And crypto prices are still wildly swinging. Perhaps we're just looking at memes to see what's going to happen. But there's also the Genius Act, which passed the Senate and is waiting to pass the House. And they're going to insert some regulation into the crypto market so this can be a stable financial game.

20:26Or not? Or are meme coins going to continue to run the show? Well, you know, as you know, usually when you say A or B, I say mix both, you know, etc. As a matter of fact, I think it was Ben Casnoka from Village who said, you know, your tagline should be nuance. It was kind of a fun observation. So one, look, one, genius stack rate. because actually in fact you know getting to more frameworks in which you know there's kind of clarity of regulatory space for entrepreneurs to act in for financial system to interact with to be clear between those is actually in fact a really good thing I think that's a very strong set of progress you know generally speaking you know kind of the pattern should be as like look what's the way that we should bring it into the financial system what are the kind of regulatory controls for whatever your particular concern is and what you'd be doing relative to that regulatory concern is the thing to be generally doing versus, you know, kind of the just hit with stick, which was one of the failure points of the previous administration.

21:33You know, I think that's a very, very good thing. And I do think that, you know, when you get to, for example, stable coins with the Genius Act and the balancing with the dollars, it's actually good for the overall dollar ecosystem. It's good for actually the world because a lot of different places, you know, can use stable coins now for financial trading systems, whether it's Venezuela or any other place. And I think that's a very good thing. But there's still a long way to go in terms of what all the use cases that people are describing, whether or not it's digital identity, you know, digitization of assets, you know, kind of scarce digital commodities, all the rest.

22:10and of course we will have lots of memes. And they may even be named after a president. Who knows? Who knows? Who knows? Perish that thought. It's a science fiction universe. So as we said, you are a philosopher entrepreneur. You got your master's in philosophy, not an MBA. And so switching to a more philosophical question. One of the things that people worry about in the age of AI is losing meaning, losing purpose, whether it's because people think there won't be jobs, so people won't have careers to give them purpose. I don't know, if we don't do the dishes every day, how are we going to find our purpose with AI creating video and content and creativity?

22:52And so what do you think about the direction that AI is heading in as it relates to sort of human creativity, flourishing, purpose? Could we go down the wrong path? Or you're convinced that we're still on the right track? A or B, please. Well, if it was Araby, as you know, from Super Agency, which I published over this year, I would say we're on the right track. Now, your setup, the question actually led me think you were going to do one of these tweets that I really loved from last year, year before, which is, I wanted AI to do the dishes so I could do poetry, not AI that did poetry so I can do the dishes more.

23:29I literally was thinking about that. Yes, it sounded like that as you were asking the question. I am ultimately very positive on a human being's ability to kind of figure out how to have a, we operate as a tribe and have a, you know, kind of meaningfulness society. Transitions can be super painful, super difficult, and we might really fuck up the transitions. And we can really fuck them up. I mean, for example, you know, take in China the Cultural Revolution, but many other circumstances where you can actually just deliberately really, really screw stuff up. And, you know, like Venezuela today, you know, like lots of good oil and totally dysfunctional society.

24:10So it's totally doable. And that's one of the reasons why they put a lot of energy in to steer and all the rest. But that being said, like we've kind of run, like if you look at a bunch of the medieval countries, including Europe, if you get to the point where like you know for example ai robots are doing all the work that's essentially like well that's what the nobility were living like when they had all the peasants and serfs and all the rest and and so you know like you have dinner parties and you know have poetry recitals and is you know bob talking to sarah and is sarah still friends with michelle and you know does michelle regard bob you know gossip is really going to increase in the age of AI.

24:50I like that take. Well, it's social. It's an interesting kind of social thing. And so, so I'm convinced, like we find meaning in how we interact with each other, what our positions in society are. Sure. Sometimes it can be, does this person get promoted, gives a higher salary, kind of classic work stuff, but it's also, is this person invited to the party, you know, et cetera. And so I think all of that, I'm not ultimately like, for example, part of the reason why people find a crisis of meaning is they go, well, this thing that I was really good at now gets outmoded. And so, for example, this thing I was really good at, which is I was a human calculator, you know, kind of keeping the accounting books and then, you know, spreadsheets come along and they're not relevant anymore.

25:32But like accounting didn't go away, right? Accounting now became other things, came, you know, scenario planning and strategic analysis. He said, well, AI can do that. It's like, okay, but maybe it's like cross-checking it or training it or managing it or running scenarios of, well, here's the, like, for example, because agents is like, okay, I think one of the things you'll see more of is, well, run five different plans and compare them with each other, right? Because which are outside of our bandwidth now, but like if you said, okay, I'm going to have this group of agents doing that, this group of agents doing that, this group of agents.

26:04Everyone is a manager. Yes, as kind of instance doing. And so I'm both bullish, a la Stay the Core Superagency, on even if we get to the Star Trek kind of science fiction, but also I also think people tend to overstate how quickly, you know, like work's going away. Um, you know, I, I, uh, teased Dario a little bit about white collar bloodbaths. It did get a lot of headlines. Yeah. Um, and, um, because I actually think that the transitions will be real and that's the thing that I think Dario was trying to emphasize, but I think human organizations and human institutions, we kind of move at the speed.

26:44We might be accelerated, might be pushed into it, but we kind of move at the speed that we need to competitively because we're kind of happy with the environment we're in. Now, challengers, entrepreneurs are always doing the new thing, the different thing, and seeing if it can stick and seeing if that can work. And that will ultimately also pick up the pace on this stuff. So a related question that we get a lot is, of course, your answer is going to be, if you're in the workplace, you have to use AI. Everyone is going to have an agent, a co-pilot, doesn't matter what profession you're in. but what are the sort of more human skills that are going to be important in the age of AI?