Hello friends 👋

I’m excited to welcome Anand Iyer to the Escape Velocity podcast, where we interview the top minds in tech and AI. Anand is Managing Partner at Canonical and Venture Partner at Lightspeed.

My co-host, CK (founder of Tensorplex Labs) points out that Anand is also a lead investor in his startup. So we’re bringing a rare, boots-on-the-ground perspective! I really enjoyed the episode.

Across a 26-year career spanning big tech, founding startups, and now early-stage investing, Anand has seen the full arc of innovation. My favourite gems were:

We wrap with a book recommendation from Anand that draws surprising parallels to venture life.

Listen on Apple Podcasts, Spotify, YouTube, and wherever you get your podcasts!

(00:00) Introduction: From Microsoft to Venture Capital Anand’s 26-year journey, learning product at Microsoft during the "Ballmer era." Transitioning to startup founder and then serendipitously to venture capital.

(12:10) The DNA of Good Founders and VCs The "temporal arbitrage" superpower for investors: predicting market trends. Why operational experience is crucial for VCs.

(24:30) AI's Impact on Startups and Operations: How AI is supercharging efficiency, accelerating product development. The "gift" of the current AI and crypto tailwinds for founders.

(36:00) Is AI Investing in a Bubble? Defining bubble behavior: 500 AI unicorns, 3 trillion in value. Concerns about "wrapper plays" and assumptions about LLM growth/scarcity.

(48:15) Investing in AI with The Founder-First Approach: Canonical's strategy as a small, specialist, early-stage fund. The paramount importance of understanding and building relationships with founders.

(59:00) Common Startup Mistakes in the First Year: The "default dead or default alive" mindset. The perils of founder ego, dropping communication, and over-hiring.

(01:12:00) The Intersection of AI and Web3 Anand's candid view is there is no product-market fit yet, but strong intuition it will get there. The "S-curve" thesis: scalable blockchains to decentralized compute, training, and inference. Virtuals and the concept of tokenized agents, Agent Commerce Protocol (ACP). The unique value Web3 brings: privacy, provenance, auditability, attribution, and self-sovereignty.

(01:30:00) Big Tech in the AI Race & Contrarian AI Views: Nvidia as a full-stack AI company, not just chips. Why Google is an interesting one to watch with its distribution and eyeballs. The challenge for Apple in the coming months. A contrarian view: Focusing on niche, smaller, hyper-personalized models at the edge, rather than just large language models and the "gold rush" for data/compute.

(01:45:00) Book Recommendation "Who Is Michael Ovitz?"

Teng Yan: Welcome back. I'm Teng Yan, founder of Chain of Thought Research, and I'm joined by my co-host, CK, founder of Tensorplex Labs. Today, we have a truly special guest whom I've been eager to speak with for a long time. CK, why don't you introduce him, and we'll dive right in.

CK: Certainly. Our guest today is Anand Iyer, managing partner of Canonical, an early-stage AI blockchain infrastructure fund, and a venture partner at Lightspeed. It’s important to mention a disclaimer: Anand is also the lead investor in my startup, so it's a pleasure to have him here.

Anand: Thank you for having me. I'm excited for this. I've been a big fan, following your research and listening to your podcast for a long time. So, I’m a long-time listener, first-time, well, joining you here.

CK: Anand, for listeners unfamiliar with your background, could you give us an overview of your journey, from starting your career at Microsoft to founding startups and eventually transitioning into venture capital?

Anand: It's been a long journey—26 years in my professional career—so I'll condense it as much as I can. I started in big tech, spending a significant portion of my formative years at Microsoft. Those years were incredibly important for me; I learned how to build products. It was also an interesting era, from 2004 to 2010, during the peak Ballmer era when we were battling open source and the iPhone, and Azure was just beginning to emerge. I learned a tremendous amount about product building within that organization, for which I'm incredibly grateful.

However, after being on the big tech path for too long, I wanted to manage my own outcome. I was always in Silicon Valley, never moved up to Redmond, and many friends in my immediate circles were starting companies, which infatuated me. In some ways, we were already an entrepreneurial organization within Microsoft, building things on the developer platform side or on Azure. So, I became a startup founder. That was the next ten years of my career. We exited our company in 2018 through an acquisition with a public company, which we then took private again in 2020. I learned a lot from that experience.

My path to venture was somewhat serendipitous, not dissimilar from the work you folks are doing now. I was really learning in public. I had been exposed to Bitcoin back in 2012, thanks to an investor, and over the years, I stayed close to the world of crypto, though not building in it directly. After my startup's acquisition, I felt it was time to move on, so I started investigating what I could build in crypto.

I had lots of ideas and initially thought I'd have another run at building a company. But it was an interesting time, and having been a founder, I had so many ideas that I initially thought we could build a venture studio to start early-stage companies. That soon catalyzed into a fund, mostly due to a lot of support from folks I’d worked with in the past and other General Partners. The adjacency was truly there, the opportunity looked great, and I thought we could bring something to the market that didn't yet exist with our funds. So, that's my very long story, somewhat condensed, into the world of venture.

CK: That's very cool. I actually started my full-time crypto career at Defiance Capital, doing both venture and liquid investing. At that time, because of how much time I spent being a user of crypto projects, I thought I'd be a pretty good VC. But in retrospect, I barely knew anything about what it takes to actually run a business. As a builder and an entrepreneur interacting with you, I can feel how having that entrepreneurial experience helps you give better advice in general. And I mean that, by the way, despite it sounding like flattery.

I'm curious: what traits do you think are needed or expected from genuinely good founders?

Anand: You were alluding to this, and to me, the rubric is quite simple: If you're seeking funding, the choice is to partner with someone who's been a career investor or someone who's been a career operator. Ideally, you get someone who's done both, and some firms excel at finding those operators to become VCs. If you can optimize for that, you suddenly have unparalleled superpowers in your corner.

The things I've learned, as I'm also relatively new to VC, having been a builder and operator for most of my life, relate to a concept called "temporal arbitrage." Investors should be good at identifying where the market is trending in the next 12 to 18 months—maybe even more condensed now, six to nine months—and getting there before consensus. Especially at the very, very early or pre-seed stages, a VC might say, "This is where we think the market is heading, and we should place a bet here." That, I believe, is a superpower, and partnering with people who have high conviction and know how to help you get there are, in my opinion, the most valuable VCs. I'm very fortunate to have worked with a few of these over the years, and I hope I can play that part for many of our founders.

We shouldn’t chase current trends or things that are already here. For example, if you look back over the last five years, people were betting on stablecoins rather than infrastructure plays, and those infrastructure plays are doing very well now. Whereas now, stablecoins are everywhere; if you're building a fintech company, you have to use stables. Five years ago, that was somewhat contrarian, and it was hard to envision we'd ever get to this point. So, that's one nuanced example, but "temporal arbitrage" is the trick to being a good VC.

CK: I agree. You mentioned how timelines have been compressed, especially in the AI era. AI is moving so quickly that all timeframes are changing quite a bit, and you really need to be hustling and reacting to all the new developments. I'm curious, in one of our regular calls, you mentioned that teams just need to behave very differently in the AI era, not just by being faster. What do you think are the biggest changes that AI has brought to startup operations and investing?

Anand: There are many vectors to this. I was just in another board meeting yesterday, and it's becoming so common for conversations to revolve around: "How do we use AI to become more efficient? How do we use it to cut costs? What are we not doing yet?" That's a question two consecutive weeks of fairly large companies have asked. Everyone wants to understand the best practices for optimizing AI usage.

Things are moving really quickly. We saw these shifts, and you folks have been in the thick of this for a long time. Perhaps at the beginning of the year, it became very obvious that if you're a really good developer—using clichés like "10x developer"—these tools can truly supercharge you. Knowing how to use these tools to accelerate your work is a superpower in itself. Especially about a year ago, it was so obvious that many of the tools existing to build products would accelerate product managers' behavior. As a PM for a long time, even as a CEO, I acted as a PM. Being able to quickly prototype something, show it to our development team, and then for them to take it from that point—it's still zero, but from zero to one, it's pretty huge. So, it's really supercharged the ability to build things and move very fast.

Obviously, with that comes some baggage at times, so you have to know how to manage it. But I'm seeing, especially in larger organizations, AI starting to infiltrate various facets, whether it's HR, legal, or obviously everyone using these tools, and productivity is going up. Every team I talk to today feels like there's a point in time where their product starts to become very important and interesting, and it's crucial to capitalize on that right now. The timing feels really great. To zoom out, AI is inflecting so heavily, the tools are there, the tailwinds are incredible in crypto and AI. I feel like this is a gift that most founders would only crave to have. Imagine where we were a few years ago. So, I feel like this is just a momentous opportunity for all of us to capitalize on, and making the most of it is our duty in some ways. That's how I think about it. It is really impacting every part of the organization.

CK: It seems to me that AI is clearly very real, but there's still a lot of skepticism around how much return you can generate from AI, especially from less technologically sophisticated investors. Maybe too many people got burned by the dot-com bubble, so they're very keen to call "tops" for tech this way. I think it's far more real than before. But the question I have is, do you think there is bubble behavior right now when it comes to AI investing? If so, where is the bubble?

Anand: Yes, definitely a bubble. There's no doubt about it. You have to stick to some amount of fundamentals when it comes to investing in venture. It's a simple truth. You have to know what's exciting, what's venture-scalable, not chase a narrative, not feel FOMO, and know when to walk away, both as a founder and as an investor. Just look at the basic stats: I think we have about 500 AI unicorns now, totaling about three trillion in value. We should ask questions like, how many of them truly have real revenue? What is the path to profitability? All the simple things we should be asking. I hope we're not forgetting to ask those questions in the spirit of just doing deals, which is tricky because that could have massive trickle-down effects all the way back to the public markets if we get burned in this process.

I'll also say this: In the last week or so, there have been conversations comparing this bubble to the dot-com bubble. I lived through that. When I moved to the Valley in 2001, it was the beginning of the crash of the dot-com bubble. I don't think history, as they say, rhymes but doesn't necessarily repeat. So there's some rhyming here, but I don't think it's exactly one-to-one. But there is a bubble for sure.

My concerns, as you asked, "where is the bubble?" There are definitely dollars flowing into the very top of the stack, pure "wrapper plays." There are strong signs of quick revenue, but a path to repeatable revenue or building a moat is hard, as the fungibility of some of these models is actually real sometimes. So, I think there's some bubble behavior there.

Aside from that, these are obvious things to me. One apparent concern is that we keep making bets, assuming that LLMs will continue to grow the same way they have been. We continue making bets, thinking that training compute will remain scarce, and that LLMs are going to be the pervasive way AI gets done in the future. We're following this classic pattern: it's expensive, it's abundant, but it might get commoditized. That's the piece I worry about a little bit. But that's also the venture game we should be playing, which is: how do we look ahead and see what things might be creeping up, given the pace of research and innovation happening here, so that we're not continuing to pour money into the same areas, but thinking where the puck might go. Yes, there are definitely signs of a bubble. Whether it will burst and how impactful it will be, how quickly we'll recover, I think that macro is actually looking quite different than it has in the past. That’s the asterisk against this in terms of the impact it could have, but it's definitely bubble behavior.

Teng Yan: If I may chime in, how do you think about investing in this kind of messy, chaotic environment? Do you try to pick the winners in each category and ensure you have a stake in them? Or is it simply very difficult, and you have to place a couple of bets and then see which ones actually turn out to be the winner in the end? I'm curious what mental model you use in this very dynamic environment.

Anand: That's a good question, and it will make sense if I share a little more about where we sit. CK has had an inside view, like being in the car as we drive this vehicle, our firm; I think he's seen some of this. If you look at the classic two-by-two quadrant of, say, fund size versus generalist/specialist, we are a small, specialist firm. We sit in a very interesting position. I have to look at that every day because if you look at that bucket, you know there are some opportunities that just don't fall into it. For example, we're not going to be chasing growth deals; we're not chasing tokens. Focusing on that makes it very interesting.

The second point is that we are looking at pre-seed and seed-stage deals, which are very, very early. It's much more art than science. So when you ask if we're trying to find a category winner or place a couple of bets, it's not really about the category; it's actually about the founders for me. That's paramount. I spend a lot of time with our founders, trying to get to know them. We invested in a team just a couple of months ago, and I'd known this team for almost a year. We spent a lot of time getting to know one another, and we had committed to investing in them, but the timing just felt right after spending almost a year with them, seeing how they operate. They also got to learn about me. Finally, we moved forward. Now, that's a very long timeframe, not normal, but this team also wasn't ready to fundraise; they weren't in fundraising mode. In fact, it was a preemptive move from my side after this whole year of "dating," to be like, "We should really do this. You folks should take some capital." It was me convincing them to formalize this relationship.

Yes, it's a slightly different motion, and I like that a lot because I want to support the teams and be a true partner. I'm not a founder, not there every single day, but how can I lend my expertise to help them grow from what is essentially a "negative one to zero" stage? A long way of saying there's a list of about 12 to 15 different ideas percolating at any given point for where we want to place our bets. Usually, if we find a team that's exploring or researching this area, that gets very exciting. We'll spend a lot of time getting to know them, and then we'll do the dance and see if it makes sense to invest. But getting to know the founders is even more paramount. I've learned simple, subtle things over the years that really work for us and for them. I've also made mistakes, by the way, and learned from them in terms of what doesn't work. But to me, the team makeup and dynamic are paramount. There are numerous examples in our portfolio where it wasn't just a "shotgun wedding"; it was more about really needing to get to know one another before deciding to do this for the long haul. That's something I've also discussed with our investors, our LPs. We're not a "quick flip" shop. There's going to be a lot of pain and suffering that comes with the first formative years of a company's operations, but at the end of the day, the reward is going to be pretty huge. So, yes, it's more traditional VC in that sense, even though we operate in crypto and blockchain infrastructure. It still looks more like traditional VC.

CK: You mentioned that the first year of building a company is very hard. What do you think are some common mistakes you see founders make in the first year, not just in AI or crypto, but in general?

Anand: Yes. Paul Graham wrote a great article—he should get all the credit for this—about "default dead or default alive." If you haven't read it, you should. This is something that invariably, it’s not exactly a mistake, but something founders forget as they get into the motion. They forget: what is the mission? What are we trying to do? Are we a revenue-driving business, or what is our North Star, and how should we measure our own success? Are we default dead or are we staying alive? If you don't ask that question, you can get really infatuated with what's happening in the market.

To add a sub-point there, another Y Combinator partner wrote about the "peaks and troughs" of company building. You go through these waves: you build your company, the PR announcement goes out, everything looks great, and you're on a high. Then you fall into a trough where you're like, "Now what?" I'm not growing, or other things are happening, we're having some friction with some folks on the team, etc. But knowing what to expect and using your advisory base around you to truly support you is key. I've warned our startup founders when they first get started that the monthly investor cadence is very strong for the first six months. Then, you hit your classic roadblocks, and the investor updates drop off. Sometimes this is because the founder's ego is too strong, thinking, "I don't want to share bad news, I shouldn't share bad news," or there's inertia. Those are situations where it becomes really hard to work with teams like this. These are some classic things I've seen.

The pattern is almost systematic; I've seen it happen way too many times. But if you mentally calibrate to the fact that "I'm actually default dead, this is not a given; we have this runway, and if we don't take off by the time we're at the end of this runway, then we're going to crash and burn," operating with that mindset is crucial. I'll credit Daniel Romero from Farcaster for this; every one of his investor updates used to start with the fact that "we are default dead, we're not there yet." And then the rest of it just followed from that because you were setting the precedent that you haven't really hit that escape velocity yet. So, that's the gist of it.

Another pattern is over-hiring. I think Sam Altman talks about this too: you tend to over-hire, perhaps because you feel like, "Hey, we've got capital, we should go do this." But it's rarely about the number of people in the first year; it's really about finding the right people. That's key. Keith Rabois has talked about this as well: sometimes you're hiring for the person, not for the role, but just find the right people around you who can get things done.

So I'd say, setting the "default dead" expectation, maintaining a consistent communication cadence (and not dropping off), chasing the North Star, iterating quickly, finding what's not working, and fixing those gaps are traditional things that will really support and help you. But I made these mistakes too. I've actually lived through this. Everything I'm saying, I've made these mistakes. I went into a hole, like an ostrich stuck my head in the sand, didn't come out, didn't talk to a lot of people. Ego was too strong. But the minute I realized I was making these mistakes is when—and there were some investors who were great at helping us through this too—those things were paramount in helping us turn the corner.

CK: Many of the things you mentioned remind me of mistakes I've made at some point, or perhaps mistakes I'm still making, I hope I'm not. One thing that truly resonates is when it comes to hiring for startups, the type of people you look for is very different from the type of people in big companies or corporates. I don't know how to explain the mentality, but if people go to work just assuming "I just need to be there and get some stuff done and call it a day," it's very hard for a startup to survive if you're full of people like this. So, yes, hiring is a very hard part of running a startup.

Anand: Yes, those first few hires. You're looking for people who are going to really do anything; it's truly part of the founding team. My favorite story is that we hired this incredible person, Steven, who was supposed to fill a very specific role: to work on our mobile apps. By the time he left, he was our CTO when we got acquired—that's how quickly he rose. There were subtle things along the way, but I remember very early on in our company's lifecycle, we were moving offices. He was that guy who showed up with his partner very early in the morning to help us pack up our stuff and put it in boxes. Him and his fiancée at the time were helping us move from one office to another. I remember thinking to myself, "I think he's always going to be there; I can call him no matter what." He's that co-founder, that person where one day he's fixing a bug, the next day he's helping us order lunch, or cleaning up. That's just the role of the early founding team: you just have to do whatever's needed at that point in time. Those are my memories too, where I look back and go, and today Steven and I are still good friends. We talk a lot. But those relationships, those bonds… it's easy to work with people like this when you feel like you're all rowing in the same direction. This doesn't mean you don't disagree—there are healthy disagreements all the time—but these are the kind of people you want to be in the trenches with.

CK: Totally. It takes time to figure out who the right people are. Once you get the right people, you gain a lot of flexibility because whatever you do, if you pivot, these people will stick around and figure out their role regardless.

Let's shift a little bit to talk about crypto, given I know you probably have the best views on the intersection of AI and Web3. How do you currently view this intersection? I remember very early in 2023-ish, when crypto people realized AI was huge, they all decided to just invest in whatever AI-related crypto projects, thinking crypto always outperforms. But in the end, I'm pretty sure most of them would have outperformed by just buying Nvidia, even a few months after ChatGPT was launched. So, I'm curious, how do you currently view the space? What is real? What is less real? What excites you the most?

Anand: It's an interesting position to be in. I live here in the Valley, and AI is kind of infused into our veins on a constant basis. We have the biggest AI companies all based here: OpenAI, Anthropic. We also have some of the biggest crypto companies based here too: Coinbase, Kraken. We've got the biggest VCs focused on AI, and the biggest VCs focused on crypto. So, it's definitely an interesting point.

One thing I'll say before going into this intersection of AI and crypto is that I want to be super clear: I don't think this intersection has found product-market fit yet. We're just not there. I invest in this space heavily and look at opportunities daily, and I'll be the first to tell you that we have not had PMF, and we should be clear about that. Anyone who says otherwise is either living in some future I'm not, or we can't see yet, or they're just fooling themselves. That's the truth. So today, we're not there.

However, I also have a strong intuition and hunch that we will get there. This makes a lot of sense. I guess what I'm roughly seeking to answer is setting the backdrop on my own mental framework and why this is exciting for me, because that's important. It's easy to get on Twitter and just say, "Hey, there's a bunch of fluff here," or "there's not a bunch of fluff here." But what are your first principles? What motivates you to look at this space? For me, it doesn't hit—by the way, I go back to my two-by-two—we're an early-stage fund focused on frontier and interesting technologies. So this does hit that rubric very, very well. If we place the right bets, as you were saying, it's good to be… I do think in bets, but when we place bets on the right people, they could have venture-style outcomes. So that's how this rubric kind of fits in.

Now, although we don't have product-market fit yet, that doesn't mean we won't. You asked me earlier about some of the things that have informed my thought process in how I work with founders. I remember this one Mark Andreessen quote, which is beautiful, and I think that's repeated maybe at least once or twice a day: he said something along the lines of, "The best founders will their perception into reality." They're actually building in markets that don't exist yet, and they're out there convincing you that this is the thing. This is where we should be placing our bets. It's where I want to spend my time. This is the future that I want all of you to come live in with me. Those are the most interesting bets to make as a venture capitalist, but also the best and most unique founders to work with, the best ones you can learn from, etc.

To me, we're in this sort of early phase of proving that this intersection has some sort of viability. There's this S-curve in my thesis that I've drawn up; they're all kind of built on top of each other. We proved that we can build scalable blockchains. We built that we can have these decentralized marketplaces for compute. And now we're moving up the stack a little bit more to prove that we can do things like truly decentralized training, like what Jansen is doing at Virtuals, working on truly proving that inference can be done through smart contracts, through these decentralized applications.

And then the next curve of this is a reality where the market is not just about the 7 billion humans, but about trillions of agents that are living on-chain and leveraging every facet of what blockchains have to offer. So that's where we are in the progression. If you look at that, suddenly you can sort of easily figure out where you think something might be fluff or noisy and where something could be real.

I’ll zoom out for a second, but when I first met the Virtuals team, they were pitching right after I’d gone to the OpenAI Dev Day in November 2022. All they were talking about was agents. That was the first time this concept of agents started to get really popular. So I came out and I was like, "Whoa, agents!" I think I put a tweet out there, and someone from the Virtuals team saw it, reached out, and said, "We're actually thinking about these things called agents." It felt too premature and too early at the time. But again, I spent several months with the team, thinking, "You folks are way ahead of where we should be. But this is one of those where if you get it right, it could be very right."

But we've still seen that many of these agents are toys; they're not really delivering on the true AI value and promise. That's the truth of it. But that's how venture works. We start off, we prove that something can be done. Most of them look like toys. And over time, we'll start to realize some monetary utility if the founding team has conviction in building and getting us there. And that's kind of where we are.

So I guess it's somewhat of an indirect answer to your question, but just to summarize, I'd say we haven't hit product-market fit yet. We're proving that a lot of things that can be done in a centralized fashion can be done in a decentralized fashion. The concern I have with those solutions a lot of times is that there's just no great reason to switch from a centralized option to a decentralized option. It's hard. The switching cost is too high, or there's no incentive and value to be able to make that switch, but we're proving that it can be done. But once those rails exist, what comes after that is, I think, where the most interesting opportunities will emerge. And that's been the same thing: we proved DeFi was a thing, and CK, you know this really well. And now it's like, as we talked about earlier, every fintech wants to dabble with stables and DeFi. So we're at that point in this infrastructure, proving it to be something bigger, but definitely not PMF yet.

Teng Yan: Yes. Since you mentioned Virtuals, I'm curious, what do you think of them now? It’s been a while. They got a lot of attention when the whole AI agent concept started to come to market. People were very excited about what these agents could do. Recently, we spent some time looking into what they were doing with their agent commerce protocol, which I think is an interesting way for them to sort of build these communication tools between different agents for this multi-agent future. I'm curious, what do you think of Virtuals at this point, if you have any thoughts?

Anand: I'll start with the team again. If you folks, and I'm sure you've interacted with them, I think very highly of this team. They've done a lot over the last several years, and there's no quitting. They're not in this for a short-term high and boost; they're in it with conviction for the long haul. There's something in their DNA that's not going to quit, and that's something I've come to respect and admire a lot. So, I'll just start with that as the basic foundation. Again, I talk about being in the trenches; this is the team that you want to be supporting and working with. Even if I wasn't an investor, I hope they win. That’s where we are with just like table stakes.

In terms of product direction, the concept of these tokenized agents got really exciting for a while, and there was a bubble forming around it. The bubble has burst. The viability of it is being questioned. So, is there a future in which any company, whether it's agentic in nature or not, obviously has an LLM backbone to it and is doing more than pure bot-like execution and has its own reasoning capabilities and can do more? For sure. Every company, as we talked about just a little while ago, AI will get infused into these organically. And so there is a world in which this starts to look a lot more interesting. The dynamic…

Again, just one more thing to zoom out: the beauty of crypto has always been that we can actually ascribe value to things that are typically very hard to ascribe value to. We went through an NFT wave, tokenized agents, right? How do you ascribe value to these? And yes, one way is you can look at how much revenue something's making and then put a multiple on it, and that's the valuation. That's how we do it in a traditional format. But what is something like this, that's so amorphous? How do you ascribe value to it? And so you marry that with some utility that's being born on-chain, which can actually create value beyond our wildest imagination. Suddenly that looks very interesting. And to me, an example of that in the last cycle, which wasn't that long ago, was AIXBT, which was kind of born on top of Virtuals.

So, I think this is a long-winded answer, but I love the team. The foundation is phenomenal. You're right, ACP (Agent Commerce Protocol) is a very interesting stab. The minute these things start to work with one another to generate value, start driving transaction volume on-chain, it starts to look very unique and different. And I think we're starting to prove that these foundation layers mean something. ACP is very nascent, but there's a lot more to come the minute… And this is a marketplace, the chicken-or-the-egg problem. The foundation now exists. We just need viable agents and then discoverability to exist for them to start working with one another. This starts to look very unique. I wrote a long time ago, but I saw this kind of parallels between what Shopify had done back in the day and what a Virtuals could do. Virtuals is following that playbook. We're going through a lull right now to find quality agents that do really interesting things. And to me, I think the big question is: does tokenization of these agents really matter? And if they don't, that's not what's important. It's that they start to transact and drive value. And at some point they can tokenize, right? It could be like a true IPO moment using the platform. But for that, when that happens, Virtuals is ready because they've got the eyeballs and they've got the liquidity to support you through their user base to actually turn the corner, raise more capital, or ascribe value to your agent. So, I wouldn't bet against this team.

CK: Very nice. In your earlier answer regarding Web3 and AI finding product-market fit, I very much agree that we should focus on things that are uniquely solvable by Web3 instead of just forcing it. It seems to me that potential ways Web3 AI can become very useful are in things like sourcing compute if compute remains a bottleneck, or proving provenance, or safeguarding the privacy of users. I'm curious, what do you think are the most important things that Web3 could bring to AI, which may lead to some big successes in the Web3-AI intersection?

Anand: To me, it also starts with the basic assumption that AI is just too centralized today. That's the reality of it. The most impressive things being built, the fact that ChatGPT's biggest user base is in India, and it's growing so quickly, becoming one of the most popular apps out there—you think about that, that's the trick. Going full-stack with this solution just made a ton of sense.

By the way, this is not a direct answer to your question, CK, but I wanted to flag that, to me, this is a formula for success. When you asked earlier, "What could Virtuals do? What do I think of them? What can any one of these AI crypto companies do?" I went slightly off on a tangent there because there is a lesson to be learned. I remember talking to GenSyn team about this about nine months ago, and GenSyn, as you folks might know, focuses on decentralized training. It's an amorphous concept to understand, but now with what they recently released with Block Assist, which is an RL (reinforcement learning) agent that's sitting with you, effectively learning about your moves and helping you get better at Minecraft using their swarm of decentralized compute—basically, it's RL happening in the background on a decentralized network, but doing a very specific use case like playing a game. They showcased in a full-stack manner what their product could do. This is unique. Crypto companies are notorious for not doing this full-stack thing, and so I'm glad that we're proving that this can be done. And we need more of that, especially in AI, because the fact that OpenAI was working on GPT for as long as they were... until ChatGPT happened, we didn't even understand what this could do. So, I think that's important.

Now, coming back to your question of where these primitives can start to make sense: Today, there's this cohort of users—and I'll start top-down—who believe in privacy. We use Signal, we use Opera or DuckDuckGo or Proton. We're a sort of outlier base, but we believe in these things because there's some semblance of how we want to operate, that we don't fully want to trust these centralized providers with everything. And we've seen this happen before with Facebook, we saw it happen with Google—leakage of user data, the use of this data can have disastrous consequences. So that's the backdrop.

Maybe just to land the plane, which is stating the obvious, we're working on truly open-source and open-weights models using decentralized primitives. There are great companies doing that today: Prime Intellect, Nous, GenSyn folks on RL specifically. And I'm sure there are several others, but that's a huge step in the right direction to prove that this can be done using a decentralized network.

The next wave of models that get built should truly be built with this sort of attribution in mind. If I'm contributing a corpus of data to pre-training, will I get credit for that? That's not really happening today. We've taken stabs with existing data models, like what Vana is doing, what Sahara has been trying to do. I think these are all kind of moving in the right direction, but we still don't see that full lifecycle happen yet. But I hope that will happen in the next wave where we have models, whether it's language models or not, that get created with these kinds of primitives as first principles, because that's something that is possible today that wasn't possible a few years ago.

So, to quickly echo what you're saying, I would say the core primitives around security, provenance, auditability, traceability, attribution, and truly self-sovereignty. These starting to come together with these primitives will be the most ideal form of AI that we can use. And I'd say again, credit to Ilya from Near, who's been parading this narrative for a long time. But I truly think this is feasible in this cycle now. However, we still need to find that ChatGPT moment. We need to find that inflection point that will make this all usable, and suddenly we'll realize that that's the thing that's going to cause us. And it's not going to come from the fact that the next AI researcher is going to leave from one lab to go to another. That's not the catalyst. The catalyst is going to be some advantage we get that we haven't realized we could even do with these technologies today. And that's the stuff that's really exciting.

Teng Yan: One observation we've heard from some people we've talked to, which I thought was interesting, and I want to get your view on that, was that you said there's a lot of noise in the AI and crypto intersection. And one of the reasons is because this intersection attracts founders who typically were not able to raise funds in the very traditional Web2 AI space because there's a lot of competition. It kind of pushes them into Web3 where maybe it might be easier to raise capital within the space because it flows pretty faster in that sense. I wonder how far do you agree with this? Do you think the breed of founders we have in this space, in your—I know you talk to a lot of founders—how do you think of this statement?

Anand: Let's see. I couldn't name many companies, maybe because they're not on my radar, so I'll caveat by saying that where the product wasn't venture-backable period, and suddenly they come into crypto and suddenly it looks venture-backable—that's a very binary thing. I haven't seen that to be the case yet, especially if there's a copy-paste of the concept, because VCs are going to be smarter to figure this out. I'm not saying there isn't "not smart money," it can happen. I haven't seen it happen yet. It goes back to VC first principles.

Before I even go to that, what I have seen happen are situations where companies that are truly sort of an AI Labs team, if they were to come into crypto, they would definitely get a very healthy multiple for sure. It's not about whether or not they'll be able to raise money; they will be able to raise money at a much higher valuation, probably even more capital than they could if they go the non-crypto VC route. I've definitely seen that happen. I'm seeing it happen right now with a team that I've gone to. And that may be just one example, but there are many such examples of this. I'm also seeing the inverse happen: I've known teams that were dabbling in the centralized space, they tried to build this in the crypto world and tried to attract more capital, but it just hasn't worked. So this is my point of view. And maybe I could be wrong about many of these, but I've seen situations that happen. I can't, and I can name those names, but I wouldn't want to do that and expose these teams this way. So I won't do that, but those teams probably know because I've talked to them from both sides.

It comes back to VC first principles and the fact that we need to suss out whether something is venture-backable or not. This doesn't mean you shouldn't build a product. Founders, if you want to shoot your shot, do it. But you don't have to be a venture-scale team. There are other forms of capital to do this. Venture isn't the only way to raise capital to build something. There are more creative ways to do this. So, just being honest with that, if the team is honest about that, that's really all that matters. As I mentioned earlier, the team is everything. I'm sitting here trying to figure out with the team what is the right time to go do something and build something really momentous. And you can also set aside for teams being like, "Hey, we're just going to launch a token and YOLO in the next six months." You could try that. That playbook is failing very quickly, and you're setting yourself up for a lot of pain if you subscribe to that. Sorry, Teng, that's not a direct answer to your question. Again, I don't think it's fair to name names here, but I hope if you stick to first principles, you'll be able to figure out what's real and what's not very quickly. And so I do see these things happen all the time. By the way, I can tell you exactly the same thing happens outside of crypto venture as well. There are a lot of things getting funded today that shouldn't get funded. That's just how VC works.

Teng Yan: Yeah, that makes sense.

CK: So, one of the feedback we've got for previous episodes is people want very specific alpha and tickers, etc. So I have the last question from my side: Among the large tech companies, is there one that you're particularly bullish on or where you have a contrarian view, and you think people are underestimating its potential, or they're giving too much trust in a certain approach?

Anand: So the question is about one large equity? Basically, around the largest AI companies like Google, Meta, Tesla, or Nvidia. Is there one that you think the market basically doesn't appreciate enough that makes you very bullish, or you just don't subscribe to the faith that the market gives to that company?

Yes, both. I mean, it's stating the obvious here, but what I've started to appreciate—and I think this is an underrated fact—is that Nvidia is truly a full-stack company. You don't think of it as just a chip manufacturer trying to solve matrix multiplication for AI; that's how we've been conditioned to think about them. But they're focused on robotics, they're building their own models—and not just generic models, but for specific industries as well, like healthcare, robotics, among others. To me, that's something worth looking into because it also speaks to the fungibility of AI at some layers here. They realized that if they can move up the stack a lot faster than someone can move down the stack. But this also talks about how OpenAI is thinking about their own hyperscalers, their own chips, and so they want to come down the stack because they've got distribution.

I would say the one to keep an eye on today would be Google. They have distribution, they've got the eyeballs. And so to me, there's—and of course, they seem to be beating many benchmarks. Although I don't use the tools, I constantly struggle to figure out how Gemini out-competes something else. But this is again, maybe a small sample size and doesn't validate what the benchmarks seem to be telling us. But that's an interesting one to watch, I'd say, because I think they are hey have been flying under the radar, but they've got such great mindshare when it comes to users using their products. Gmail is pervasively used, and so is Google.com, among others. There’s something there that’s worth tracking and looking into.

Obviously, Apple has been asleep, and so we’ll see how that plays out, whether it’s through acquisitions or... I mean, they’ve got further distribution, which has always been a strong suit. The next few months are going to be really challenging for Apple in terms of what they do here, because they have new devices coming out, which is usually an inflection point for them. It's been slow to see that catalyst kind of play out. I don’t think... I think I’m just repeating everything that you’ve probably already seen on various other sites, among other things.

However, a trend that I think is worth paying attention to—and this isn't about going long or going short—but I think it’s worth looking at today is: We are so fixated on LLMs. We’re so fixated on expanding these large models to collect all the data and become the Google of this era. That's what many of us are trying to chase. And we’re proving—and ChatGPT and maybe Claude have proven—that there’s fungibility when it comes to these models. Distribution is paramount; that’s how you’re going to hit true escape velocity.

I think a trend that's worth poking and looking into is more niche, smaller models. They may not be language models, but they are smaller models. CK and I have talked about this in the past in other meetings, but I think it’s a contrarian view to where a lot of the attraction is going.

So, focusing on the edge, number one, and on truly small, focused, hyper-personalized models is really interesting. This kind of defeats the narrative of the gold rush for data, for compute, for chips. I believe this will change; I'm not saying that market will shrink, but I think this other market will grow in the next few years, and that's really exciting to me. There's no clear winner here yet. There's a lot of research happening in this space, but I think that's really exciting. And that's actually where I've been spending a lot of our time, looking at these sorts of opportunities.

But yeah, that's one where I try to keep an eye out.

CK: Appreciate the answer.

Teng Yan: Amazing. That’s awesome.

I think we'll just wrap the part with more of a fun question. It doesn't have to be anything to do with AI, but I was wondering, is there any good book, movie, or show you've seen recently that you would recommend to our audience? We're always looking for this kind of alpha.

Anand: I'm reading this book right now: Who is Michael Ovitz? If you guys have heard about Michael Ovitz, this is again something that Marc Andreessen had flagged a long time ago. Michael Ovitz is an incredible agent for movie stars. But there are a lot of parallels here in terms of how you run a VC firm and how you recruit and retain talent on the movies side of the house, let me put it that way.

I'm reading that because we're going through an interesting inflection point. There's a lot of great talent out there. How I can learn to become the best support system to attract and retain talent, like Michael Ovitz has done, is something that I think is... Again, there's such a depth in the way he writes and the things that he's gone through, so I'm kind of re-reading that. It's not for everyone, but it’s also a quick read. It's a thick book, but it's a quick read in my opinion. So that's what I'm going through today.

CK: Yeah.

Teng Yan: Awesome. Yeah, we'll add it to the show notes. And definitely, I'm always on the lookout for good books, so we'll add that to the reading list also.

But this is great. Yeah, thanks so much for taking the time to have this chat with us on AI. I think we can call it a day for this.

Anand: Awesome. Thank you, guys. Thank you, guys.

See you in our next podcast!

Cheers,

Teng Yan