YT YouTube2Text

YouTube2Text → Transcripts → The Agent Economy: Why AI Needs Crypto - TOKEN2049 Singapore 2026

The Agent Economy: Why AI Needs Crypto - TOKEN2049 Singapore 2026 — Transcript

by TOKEN2049 · 6,278 words · 951 segments · language en · Watch on YouTube

Open in transcriber Summarize with AI Download SRT/VTT

Full transcript

0:08All right, so I will say I have both
0:10been on this panel 3 years in a row
0:13probably with Ilya.
0:15And this is the first time I think I'm
0:17actually positive on the answer to this
0:19question.
0:20Uh, but uh, you know, we have uh
0:24two great panelists. Uh, they both have
0:27very different but complimentary views I
0:30imagine on AI but we'll find out
0:33a little more. Um,
0:34so maybe, you know,
0:36you guys both have long stories in
0:39crypto,
0:40have worked at different entities, you
0:42know, starting a protocol,
0:44creating a a roll up at a large
0:46exchange.
0:48But
0:49you know,
0:50not everyone knows what your your views
0:52on crypto and AI AI are, what you're
0:54sort of how you've kind of interacted
0:56with the fields, how you've been part of
0:58it. So maybe give a little background uh
1:00about, you know,
1:02where you are with your crypto and AI
1:04journey.
1:05Ilya's
1:06begins before he started NEAR and
1:08probably so.
1:09>> Yeah, so I mean, obviously my AI before
1:12crypto uh has been working on the
1:14transformer architecture at Google
1:16research and then starting NEAR as the
1:19AI company uh back in 2017. Uh, we were
1:22a little bit too early uh to effectively
1:25write coding.
1:26And then realizing that we actually need
1:29a blockchain protocol so we can
1:31coordinate people, data, and over time
1:34more things in a decentralized kind of
1:37globally accessible way. And so started
1:39NEAR protocol in 2018. But over past, I
1:42guess, like 2 years been focused on on
1:45how do we
1:46A make sure that AI stays private,
1:49verifiable, and, you know, anti-swarm
1:53uh
1:54uh detection and anything like that. And
1:56at the same time, how do we actually
1:58leverage kind of blockchain markets
2:00economy to
2:02really unlock I call it right now AI as
2:05a single-player mode. And so, how do we
2:08unlock this multiplayer mode by using
2:10blockchain as a kind of back end to AI
2:12infrastructure?
2:14>> Yeah, and I'm Jesse, a less illustrious
2:17AI past than Ilya, but
2:21yeah, I've gotten to watch Base grow as
2:24a kind of platform for AI agents agents
2:27tokenizing themselves, buying inference,
2:30genetic payments.
2:31It's been cool to see that develop over
2:33the last couple years.
2:34And then personally a deep connoisseur
2:36of being able to write code again after
2:395 years of not writing code and just
2:41getting to plug back into the matrix.
2:44Which has honestly been one of my
2:46favorite things about this year where it
2:48feels like we totally crossed the chasm
2:50where, you know, a year ago I was trying
2:52to write code it was still this
2:54like starting stop thing and today it's
2:56like all day every day I'm writing code.
2:58In between meetings and that experience
3:00I think has been
3:02I think one that many of us have been
3:04through and we're reaching this
3:05convergence point where something's
3:07going to happen. And it's happening, but
3:10we're we're not quite sure exactly how
3:12it plays out and
3:14my bet is that AI and crypto are going
3:16to come together in some really profound
3:18way to make that thing happen and and
3:20make it a better thing for the world.
3:22So, I'm excited.
3:23>> Yeah, actually I I remember and I met in
3:262018 and he was like, I have a program
3:28synthesis company, which is what Vibe
3:30Coding was called prior, you know, in
3:32the the prior dark ages.
3:35So, you know, you both talked a little
3:37bit about AI and crypto and and sort of
3:40different ways you think there might be
3:42symbiosis.
3:44I think historically people have been
3:45sort of pessimistic, maybe didn't think
3:49there was, you know, I've I've been on
3:50the
3:50>> not people you Hey, hey, hey.
3:52>> [laughter]
3:53>> Not just me, though. I've been on like I
3:54said, you and I have been doing this
3:56panel since 2023. And I don't think
3:58people were universally positive,
4:00whereas now I think the vibes have
4:02changed a lot. So, I guess let's just
4:04start with like what what are you most
4:06optimistic about the convergence and
4:08convergence and symbiosis of AI and
4:10crypto?
4:11>> So, I think there there's things that I
4:13hear right now, right? Which is how do
4:15we make sure we have kind of universal
4:18access to AI? How do we make sure when
4:20you use AI, it's actually the thing you
4:22want to use, not somebody inserting like
4:24malicious tool outputs into your agent?
4:28Like that's available right now, right?
4:30We have Near AI, we have a bunch of
4:31other providers. We have obviously
4:33Venice and kind of few other places
4:36where you can now tap into. And you can
4:38pay with Zcash, you can pay with
4:39Bitcoin, you can pay with kind of any
4:41crypto. And I think like that's kind of
4:43now universally like here.
4:46The cool thing I'm really looking
4:48forward for the next year is And I And I
4:51And I use this example because like
4:52people like, "Oh, you know, your agent
4:54will be able to do your groceries." And
4:56it's like the reality is the consumer
4:58apps are so well optimized that like the
5:01the change in experience is not that
5:02much. What really is a suboptimal
5:06experience is all of the business
5:07commerce. Is the finding suppliers,
5:11doing billing and invoicing, making sure
5:13legal contracts are redlined because you
5:16actually have a very trust kind of lack
5:19of trust environment in business
5:21transactions. And this is where I think
5:23AI and blockchain really coming in
5:25together. Blockchain provides us this
5:27trust infrastructure, escrow, payments,
5:29derivatives, all of the things we can
5:31do. And AI does all of the kind of
5:34emailing, heavy lifting to do actual
5:36discovery of products, and do some
5:38facilitation. And so, I think that's
5:40actually when we see, you know,
5:42trillions of actual economy moving on
5:44chain, not just speculative, not just
5:46kind of assets, but actually doing real
5:48commerce work, you know, shipping, you
5:51know, steel and
5:52doing like chemical supply chain. All of
5:55that things starting to really kind of
5:57move to blockchain as this AI and crypto
5:59coming in.
6:00>> Yeah, I think that's going to happen. I
6:02also do think it's going to come on the
6:04consumer side, particularly consumer
6:06finance intersecting with AI. I mean, I
6:08had the personal experience where I saw
6:10someone just a few weeks ago tweet about
6:12how they were using something called
6:13bean count, which is like open source
6:16ledger. And I was like
6:18plug in all of my I said this to my
6:20agent, plug in all of my accounts, put
6:22it into this ledger, and then like give
6:24me my books and my P&L and everything
6:26that's happening. And the agent was able
6:28to just do it. And that was kind of a
6:30zero for one moment where I didn't have
6:32any of that visibility into my personal
6:33finances, and then I did. And I think
6:36money in general for so many people is
6:38this confusing thing where they maybe
6:40don't know how to invest, they don't
6:42know how to budget, and it's often felt
6:44like oh, I need to learn and it's going
6:47to require reading books or going to
6:48school or whatever it is. And now we
6:50have these insanely powerful
6:53agents that I think we can plug into our
6:55financial lives that are going to be
6:56able to manage our money, trade on our
6:59behalf, and really simplify how we all
7:02interface with money. And so for me,
7:04that's this just super exciting thing
7:06that I feel like we're still just
7:07scratching the surface of. So I'd say
7:09that's my take on the kind of like
7:10consumer
7:12utility of AI and agent and crypto. Now,
7:16on on the maybe more philosophical side,
7:19I do feel like
7:20you know, you have this kind of like
7:22massive explosion of of content being
7:25generated with AI and like what's real.
7:28And there is this convergence with
7:30crypto where crypto provides the
7:31hardness. It's like the verification
7:33layer. It's the infrastructure that
7:35feels almost perfectly suited to kind of
7:38counterbalance the explosion of
7:41everything that's happening with AI. And
7:43so, again, it feels like they're
7:44converging in a way that's going to be
7:46pretty profound in terms of how our
7:48world is shaped over the next couple
7:49years.
7:50>> Yeah, I would say that generally when
7:52you talk to any people who working in
7:54AI, they usually have like this like
7:56single player mode. And we've been
7:58trained in this like adversarial
8:00multiplayer, you know, kind of aligning
8:02incentives. And it feels like, yeah,
8:05bringing these things two together,
8:06right? The intelligence of AI and the
8:09multiplayer adversarial together. And
8:11you've been writing a lot about just
8:13like applying the same primitives to AI
8:15inference markets and kind of just
8:17generally compute markets. I think that
8:19that's like the missing piece of in many
8:21ways for AI thinking.
8:23>> I mean, yeah, to your point, a lot of
8:26stuff that's been coming out sounds like
8:28crypto. So, like a lot of stuff on the
8:30consumer app side,
8:32they're just been things about like
8:33security protocols for consumer agents.
8:36And you read the the docs and it's like,
8:38wait, you guys just made a pseudo
8:39blockchain. I don't like what you're
8:40talking about the being like I think AI
8:42people actually rediscovering blockchain
8:44stuff a lot and they don't realize it
8:47because they hate the word blockchain.
8:48It's like a bad It's like a, you know,
8:50bad word for them.
8:52But I think, you know, if let's take a
8:53little look back at the last few years
8:57of of crypto and AI. I think we had a
8:59few phases, right? The first phases were
9:01focused on
9:03rent a GPU, like trying to kind of like
9:06a little too skeuomorphically to Bitcoin
9:08be like, I can get a node. Here's a node
9:10for you. We have a decentralized network
9:12of nodes.
9:13One will be available. The interesting
9:15thing about that was crypto people got
9:17the market right. There's this huge
9:18market for inference providers that, you
9:21know, is a tens of to hundreds of
9:23billions dollars a year market right
9:24now.
9:26It's just that somehow that adding a
9:27token to it didn't make it a better
9:29product, right? Uh after that we fell
9:31out in like decentralized training or AI
9:34agents. In fact, AI trading agents were
9:36probably just too early, like the AI XBT
9:39and and things like that.
9:40So, what do you think if we look back
9:42that crypto founders got wrong, you
9:45know,
9:46because like it seems like they've
9:47actually been kind of early at the right
9:49idea, but they kind of got the wrong
9:51framing or the wrong go-to-market and
9:54and and what sort of advice
9:56from your perspective would you give
9:58kind of founders to?
9:58>> I think what what happened is the crypto
10:01people who actually wanted to do
10:03AI properly just went and did AI
10:06company, right? You have Alex who went
10:08and did Open Router.
10:10And uh he's like, "Okay, we don't need
10:12crypto. We can accept crypto, but we
10:14don't need crypto underneath, right?"
10:16And as you said, right, they used the
10:17same principles. Same as like all the
10:20miners, like Corvive is effectively
10:22ex-Bitcoin miners, right? And like now,
10:25you know, half of the neo clouds are
10:26effectively miners.
10:29Um and so you kind of see that across
10:30like few different uh
10:33like parts of the stack where when
10:35people like, "Okay, we need to go do
10:37it." They kind of like, "Okay,
10:39all the blockchain stuff is kind of
10:40slowing us down and like really
10:42limiting." And I think there is few
10:45reasons conceptually for that, right?
10:47One is where we normally say like, "Oh,
10:50you know, the token is a great incentive
10:53to like bootstrap network effects and do
10:55this." I would say like past 2 years
10:57that hasn't been really the case. I
10:59think the kind of new token meta hasn't
11:01really played out as well.
11:03And then the reality is like token is a
11:06product on itself, and so you're kind of
11:08like trying to build token and actually
11:11AI product at the same time. And, you
11:14know, building good products in AI is
11:15already hard. So, I I think it's also
11:17just like split at the focus. Um I think
11:20now that
11:23like some of the companies may actually
11:25like use a token when they already
11:27established their product and they can
11:29use it to bootstrap the network effects
11:32as well as the market I would say
11:34turning. Like I think some of this is
11:36just like literally the timing where,
11:38you know, like the AI AI 16Z and things
11:42like that like launching a token first
11:44and then trying to build a product, like
11:45that's a really bad, you know,
11:47sequencing, right? We which is they
11:49could have built effectively like
11:52product first and then launch a token to
11:53like bootstrap the the economy. So I
11:56think that that's, I mean,
11:58kind of historically been one of the
11:59challenges.
12:00>> Yeah, and I think that last point you
12:03made is almost like a universal rule.
12:05That's very hard to
12:07um have a early-stage product and launch
12:09a token very early and make that work
12:12just because there's so much iteration
12:13that you have to do on the product and
12:15the second you have to also manage this
12:17economic system. I think it just adds
12:19another layer of hard mode. There's some
12:21folks who can do it like I I look at
12:22like Venice. They launched it relatively
12:25early in the life cycle, but you have
12:27Eric leading Venice and like he's he's
12:29had a decade of experience navigating
12:31those waters.
12:33I I do think that like I'm not even sure
12:35that's necessarily that people got
12:36things wrong or that like the past eras
12:39like they're mistakes.
12:41I think a lot of it's just timing. Like
12:42these things take a really long time and
12:44the crypto systems if you look at how
12:46much they've matured in the last 5 years
12:48where we now have stable coins, we now
12:50have tokenization of equities and all
12:52these other assets.
12:53Um the chains are fast and cheap and
12:55easy to use. Like that infrastructure
12:57didn't exist 4 or 5 years ago. And the
13:00the other side in the AI side like the
13:02models weren't at the point where um
13:06you know, I think they were mature
13:07enough to really make up for the crypto
13:09systems. And so, you know, could crypto
13:11people maybe have invented OpenAI or
13:13Anthropic? Sure, but I think like my
13:16guess is that again they're just going
13:17to converge and that the crypto stuff
13:19will have matured, the AI stuff will
13:21have matured and over the next few years
13:23um we'll start to see more and more of
13:25it coming together. And it's happening,
13:27right? Like you look at Venice and it's
13:29like it's private AI that's a consumer
13:31experience that has a token built into
13:33it that's working. It's working for the
13:35business, it's working for the the
13:37asset, um and I think it's working from
13:39a agent and AI perspective. And so, um
13:42I'm pretty optimistic that that that
13:43kind of flywheel is going to keep
13:44spinning and people are going to keep
13:46finding those intersections.
13:47>> Yeah, so I think you bring up a a great
13:49point with talking about some of the
13:51successes that we've seen, um especially
13:54recently this year, right? So,
13:56you know,
13:57I think
13:58a product like Venice is also sort of
14:01a kind of a consequence of open models
14:03becoming better. And the sort of open
14:05model ecosystem is a lot closer in ethos
14:08to crypto
14:09and sort of the genesis of it looks much
14:11more similar. So,
14:13what do you think people have gotten
14:14right? And like why has it flipped so
14:16much in the last 12 months where crypto
14:19and AI projects have been able to get a
14:21lot more mainstream success, um you
14:23know, everyone from Noose with Hermes,
14:26Near, Venice, etc. Some had tokens, some
14:29didn't have tokens, but they all kind of
14:31start from the same place. What what
14:32were the biggest drivers of success do
14:34you think for for these projects? And
14:37sort of where do you see them going for
14:38the next 12 months? So so next time Ilya
14:41and I are on this panel, we have a new
14:43answer.
14:44>> Great product.
14:46I don't know, that seems like the
14:47consistent theme. I've had my Hermes
14:49agent for like 6 months and I love it. I
14:51use Venice and it's a great product.
14:54Near clearly has found product market
14:56fit on the intent side. And so, again,
14:58like crypto as a economic model and as a
15:01incentive tool and as a way of
15:03capitalizing businesses, um I think
15:06obviously a superpower, but you still
15:07need the great products. And I feel like
15:09generally we've turned the corner as an
15:11industry where that's what we're
15:12laser-focused on. That's what everyone's
15:14being evaluated on and that means that
15:16unsurprisingly we're starting to put up
15:18more wins.
15:21>> Yeah, I agree. I think
15:22focusing on the product and then at the
15:25same time the catching the wave of
15:28effectively I think what happened
15:32that open weight models gotten way
15:34better, right? Way closer to frontier
15:37from a distance, but also just generally
15:39the level now, you know, I mean the
15:41example we use is like we run deep seek
15:45flash for formal verification and we
15:48beat everybody else, right? Including
15:50GPT soul, you know, Claude uh and like
15:54custom models that people with hundreds
15:56of millions of dollars trained, right?
15:57So like the models are actually really
15:58good, they're cheap and you can run them
16:00for longer because of this. And so you
16:02can run agents, you can run, you know,
16:05kind of private compute and you're
16:06actually getting uh real results now.
16:09And so I think like that was a big
16:11movement. Obviously Hermes caught on
16:14kind of this personalized agent movement
16:16as well. So I think I mean as usually
16:18with any startup, right? Like the
16:20category definition is, you know, you
16:22you build a great product and you caught
16:24a kind of some wave that you're able to
16:27capitalize by being there and really
16:30providing that great product to the
16:31expanding audience.
16:33>> Yeah, I think one thing though is that
16:36the crypto founders have been able to do
16:38a lot better with
16:39kind of being able to mesh together
16:42different aspects of of AI products.
16:44Whereas like I think maybe a year ago
16:46the open source ecosystem wasn't very
16:48good. The kind of ability to to go from
16:51consumer product down all the way to
16:54like model that's performance, harness
16:56that's performance, etc. wasn't
16:57available.
16:59So I think
17:00it looking a bit forward, you know,
17:02I think as these models have gotten more
17:05ingrained in people's lives, people have
17:06started valuing the privacy aspect a lot
17:08more, right? And so that's where you're
17:10starting to see a lot more of the
17:11overlap with crypto. So I'm curious, how
17:13do you see pri- the demand for private
17:16compute, private inference, private AI
17:18evolving and how where does that
17:20intersect with crypto? Because both of
17:21you have sort of brought up the privacy
17:23aspect in different ways and are sort of
17:26more closely involved with that. So so
17:27how do you see that evolving?
17:29>> Yeah, I mean I I would say like so we
17:31launched our private offering December
17:33last year and kind of when we started
17:36like offering and selling it, the the
17:39market wasn't ready, right? A lot of
17:41people were using frontier models. So
17:44you go to businesses they're like, yeah,
17:46we're fine, we're experimenting, it's
17:48not critical.
17:49And I think
17:51even last three, four months effectively
17:53the page has completely turned. People
17:55are like, okay, well, I'm actually
17:57feeding all of my company data into
17:59these AI models and so anyone who has
18:02access to these logs can reconstruct
18:04literally the company from it.
18:06It is making critical decisions, so I do
18:08want to know what exactly is happening,
18:10right? I want a way to kind of inspect
18:13and and
18:14like know the exact model, right? not
18:16you know, Astra first week was great,
18:18second week became you know, way dumber,
18:20right? So like I don't want that to
18:22happen. And then
18:24from even consumer perspective, right?
18:26People are like, oh, actually I'm also
18:28feeding more and more of my financial
18:29data, my medical data, my all of these
18:32decisions are now there. So it's it's
18:34also important that, you know, this is
18:36not going somewhere. And obviously every
18:39single time there's a open AI news that,
18:42you know, they potentially trained on
18:45some
18:46uh you know, data that they were not
18:48supposed to and effectively used it to
18:51to neighbor stocks or uh kind of example
18:54with Anthropic launching Claude design
18:57after being a Figma partner. Like every
18:59single like that is not helping and
19:02obviously them coming out and saying
19:03they're going to kill, you know, there's
19:0410% chance everyone. So like they're
19:07kind of also undermining their own
19:08position by doing that and so people
19:11want more and more like, hey, we
19:12actually want to be in control." And so,
19:14and that is literally what crypto been
19:17kind of like from a philosophical
19:19perspective, "Hey, this is sovereign.
19:20This is control. You You get kind of
19:23privacy, verifiability, and access." And
19:25so, yeah, I think it's been massive
19:27transition to this just like across all
19:30of the things that are happening in the
19:31market.
19:32>> Yeah, and I think we've seen that
19:34firsthand internally at Coinbase
19:36actually, where um if you look at the
19:38trajectory of our usage, you know,
19:40Coinbase is a big company. I think we we
19:42push pretty far out on the kind of like
19:44overall AI adoption. I'd say at the
19:47beginning of the year, um we were all in
19:50on the frontier models. Um we were
19:52pretty locked in on one provider that
19:54basically everyone was using. And then
19:56actually a credit to Brian, I think kind
19:58of end of Q1 of this year, he basically
20:01said, "We need to not be locked into one
20:03model or one one provider, and we need
20:06to figure out how to use open-source
20:07models, and we need to get into a place
20:09where we control our data." And over the
20:11last 6 months, what's happened is we've
20:13basically massively diversified, where
20:15now we run our own harness, we are
20:17routing across some frontier models, but
20:19then a ton of open-source models that
20:21are available to us, and we're being
20:23much more conscious about where all of
20:24the data flows in the business. And if
20:26you look at overall token consumption,
20:29it's like up into the right. But if you
20:31look at costs, we've basically been able
20:33to kind of like get it to tail off and
20:35flatten. And I think from a productivity
20:37perspective, we've made the team way,
20:39way, way more productive and and
20:41successful with AI than we were 6 months
20:43or 9 months ago. And so, my sense is
20:45that that same trajectory will probably
20:48come to every business. It will come for
20:49open-source. It will come for privacy.
20:52It will come for control and cost
20:54management. I think it will also come on
20:56consumers, but they'll probably be a
20:58little bit further behind. Although it's
21:00obviously happening to some extent, but
21:03yeah, like people want sovereignty.
21:05That's why we're all in it. I think
21:06crypto people are a little bit earlier
21:09on that curve, but I expect it to go
21:11everywhere. And one thing I wanted to
21:12just kind of separate from the privacy
21:14point, but I wanted to react to as you
21:16said this thing about crypto founders
21:18are a kind of uniquely good at figuring
21:20out how to work across the stack. One of
21:22the things that I'm most excited about
21:23is is kind of like a part of Venice's
21:26product stack, which is what they've
21:27done with DM, where they basically said,
21:30"Hey, we have this consumer product, but
21:32let's figure out how do we like
21:33literally tokenize the inference and put
21:35it on chain and then start building a
21:37market around it. And I think that's
21:38such a great example of figuring out,
21:40"Okay, we built a great product. Now,
21:43how do we leverage the crypto thing to
21:45make it so that product can be more like
21:47a platform, so that can be more
21:49composable, and so that honestly we can
21:50start to invent new markets that
21:52increase access, give people more
21:54control over their underlying inference,
21:56and again, increase sovereignty of the
21:59way we're all using AI." And so, my hope
22:01for how we kind of see crypto and AI
22:05evolve over the next few years is that
22:07we are the people who are like
22:10steadfast fighting the fight to be like,
22:12"Yeah, we're not going to just have a
22:14few labs controlling the future. We're
22:16going to have open economies. We're
22:18going to have open source. We're going
22:19to have open systems." And that's going
22:21to be the future we build on top of this
22:23new platform.
22:24>> Yeah, so actually I'll get back
22:26[clears throat] a little bit to the
22:27inference market side and how that's
22:29that's kind of grown. But actually, you
22:31made me think about one question that
22:34you know, there's a lot of things crypto
22:36people were a little too early to. One
22:38thing about kind of the Web3 movement in
22:42you know, 2019 through 2022, there's a
22:45big focus on like owning your own data,
22:48right? Like you get compensated for your
22:49own data, anything you're kind of
22:51providing
22:52kind of
22:53gets gets at that.
22:55In the current AI world,
22:57every frontier lab, every open even the
23:01open source companies are
23:03mining your traces to improve their
23:06model, to improve the harnesses, to
23:07improve And there's no real actual
23:10compensation. In some sense, a lot of
23:12the original web 3 ideas
23:15about data ownership, data provenance,
23:17and sovereignty around that should apply
23:19here. So, do you think that will be that
23:22will happen in the in the next few
23:24years, or do you think it's it's still
23:25too far out? Crypto people were just a
23:27decade early versus a few years early.
23:30>> So, we actually have a product in
23:32progress. We got we got to tease it, but
23:35it's effectively you I mean, nearly I
23:38provide you private inference, but if
23:41you do if you want, you can actually
23:43contribute your traces back and get paid
23:47either in more inference or in in some
23:49other way. Uh and effectively, that data
23:52would be then used to uh train models.
23:55And so, the interesting thing is
23:56actually like figuring out how to value
23:58that data. Because one of the problems I
24:01think kind of when we were talking
24:02about, you know, own your data before
24:04has been we haven't really like had a
24:08good way to evaluate how much
24:11like how valuable this data is, right?
24:13And it's like it's just a bunch of
24:14random, you know, articles or bunch of,
24:17[snorts] you know, documents. This can
24:19be anything from worth nothing to, you
24:22know, this is a internal Coinbase
24:24documents and like highly valuable, but
24:26also shouldn't be uh shared. And so, I
24:29think [clears throat] now with kind of
24:31agentic traces, there's is a lot clearer
24:34way to evaluate this because you
24:36effectively have human input to like
24:39course correct the model.
24:41And that is extremely valuable and and
24:43and can be evaluated uh kind of
24:45programmatically. And so, uh so I think
24:48like we'll we'll see this, but it needs
24:50to be combined with effectively
24:52verifiable inference, so you know that
24:54the AI output is coming from indeed the
24:58the So you kind of needed like to build
25:00the stack up to get to this level of of
25:04kind of contribute like data evaluation
25:07and and uh
25:09like valuing it.
25:12>> I think it's I think the costs are
25:13coming for the labs. I'm not sure it's
25:15going to start in the consumer side. One
25:18interesting trend that we're seeing is
25:20that we we've worked with AWS and
25:23Cloudflare who both independently
25:26um have basically said, "Hey, we have
25:27all of these businesses that are running
25:30web businesses where they're being
25:32incessantly scraped by labs and agents
25:35and they're losing monetization as a
25:38result of it and we don't know what to
25:39do about it." And the solution that
25:41they've come back with is they've said,
25:43"Hey, let's empower these businesses to
25:46charge on a per page basis for that
25:48content." And so we've worked with them
25:50to basically make it so like if you run
25:52a website on Cloudflare AWS, you can
25:54check a box, you can set per page
25:56pricing. It can be like 1 1 millionth of
25:59a cent per load. It gets charged with
26:01X4R2, you settle in USDC on base, and
26:04now there's a business model. So that
26:06news site can charge a 1 1 thousandth or
26:091 1 millionth of a cent and it adds up
26:11because the agents are mining it or the
26:13labs are mining it. My sense is that
26:16what what may happen is that over the
26:18next year you basically see this shift
26:20where the labs have been able to get a
26:22lot of free data from websites and the
26:25internet over the last 5 to 10 years and
26:28all of those websites are basically
26:29going to get hip to it and be like, "No,
26:31no, no, no, no. Like you're going to pay
26:33us for our data." And then at the same
26:35time you'll basically see that kind of
26:38increase the value of the underlying
26:40data that's being fed into it. And then
26:42consumers are going to realize where
26:43it's like, "Oh, if all these businesses
26:45are monetizing and this is a narrative
26:46about how they're monetizing, that's
26:48buoying their stock stock prices, why
26:50not me?" And then we'll see the exact
26:52same thing happen on the consumer side.
26:54I don't see any reason why there
26:55shouldn't be like the products you guys
26:57are building where it's like, yeah, you
26:58can trade on my data, but you're going
27:00to pay me for it. Like I, you know,
27:03I'll Yeah. Yeah. I think there's a ton
27:05of opportunity there.
27:06>> Well, I think it's interesting both of
27:08you highlighted something
27:10funny, which is, you know, I think about
27:12the Web3 era, I think about 2021. It was
27:14really focused on like consumer personal
27:18stuff, right? Like the internet for the
27:20consumer that has sovereignty over their
27:22data and life.
27:24But in both of your examples, you
27:26focused on businesses sovereignty. Like
27:28the businesses care more about valuing
27:30their sovereignty than consumers. Which
27:33is
27:33kind of a evolution of Web3 in some ways
27:36to this market, which I think is is
27:38something that's made the crypto aspects
27:41of the current AI market much more
27:43amenable.
27:44>> And I think it's just indicative of
27:45where we are as an industry. Like if I
27:47look at what we're seeing from a product
27:49market perspective, it's basically like
27:51trading, payments, and financing have
27:54hit mainstream
27:56adoption among businesses where it's
27:57just like easier, faster, cheaper, and
28:01better for them to run their business on
28:02this new rail than the old rail. And so
28:04that's leading to like businesses moving
28:07over. And then the consumer still
28:08benefit because they get better
28:10products, but it's no longer like I as a
28:13consumer need to be a crypto person.
28:15It's like I as a business can choose
28:17this better technology to let me run my
28:19business better, and then that's going
28:21to benefit all of my customers. Which
28:23honestly is a pretty cool change. I
28:24think it's definitely led maybe to a
28:26little
28:27turmoil in terms of crypto's like
28:29emotional state over the last couple of
28:31years. And identity, yeah. And identity
28:33cuz I think we all are pretty used to
28:35like selling people the crypto thing,
28:37but now we're no longer selling the
28:38crypto thing, we're just selling better
28:40infrastructure to help people build
28:42better products. So, I don't know. I
28:44think it's cool.
28:46Change though.
28:47>> I mean, I I think it's like it we see
28:49this with Venice as well, right? There
28:50is a consumer story as well. It's just
28:52like you really need to build a product
28:53that's like actually matches consumer
28:56like level of experience, right? I mean,
28:59this is what we're doing with near.com.
29:01Like it's fully chain abstracted.
29:02There's no gas. There's no chains.
29:04There's no like kind of complicated
29:07things. Like it needs to just work. And
29:09when you hit that bar, then like
29:11consumer adoption comes and then you can
29:14educate them what the value of privacy,
29:16what the value of sovereignty is, and
29:18how that enables them to do more things.
29:20But I think crypto has struggled to get
29:23to that level experience for the most
29:25part. And like there's a lot of like
29:27infrastructure challenges that were
29:28there, etc. With businesses it's easier
29:30because as soon as effectively
29:32regulation lifted, it's like, "Oh, it's
29:34going to be like 100 times cheaper for
29:36me to do this, you know, payments thing.
29:38Yes, we're going to do it, right?" So
29:39like the adoption there is very like
29:42calculated, you know, spreadsheet based
29:44than uh kind of oh not always a product
29:47kind of experience based.
29:48>> For sure. But, you know, saying web 3
29:50for businesses does sound a little like
29:52corporate slop. [laughter]
29:54It's like, you know, kind of sad.
29:56Um
29:57>> It's sovereignty for businesses.
29:59>> Sovereignty for Yeah, that's a little
30:01bit better.
30:01>> [laughter]
30:02>> Yeah, it it might need a little
30:03rebranding.
30:05Um all right. So, we got 2 minutes left.
30:08You know, you both have talked a lot
30:09about new marketplaces that have evolved
30:13coming out of mixing crypto and and web
30:163. I think uh
30:17I have
30:18certainly have been thinking a lot about
30:20this lately and and writing about how
30:22the open inference marketplaces, as well
30:25as sort of harnesses holding the user
30:27data,
30:28sort of looks a lot like crypto. You're
30:30unbundling the frontier labs. The
30:31harness stores user data. It's like a
30:33wallet. The router finds you where
30:36you're supposed to go. The models are
30:38like public goods. And then the the
30:40inference writers are like, you know,
30:41liquidity providers. So, I think
30:44now that we've seen that crypto-like
30:46market structures are showing up in AI,
30:48where do you think the next year, like
30:52the next big new market structure will
30:54show up? It, you know, we have
30:55inference, we have data,
30:57you know, we kind of have like other
30:59forms of user monetization. You know, as
31:01you both sit in your seats of of of
31:03different types of uses, where do you
31:04feel like the the next big new
31:06marketplace in crypto and AI will come
31:08be?
31:08>> Well, well, the the biggest one is the
31:10GPUs themselves, because the the market
31:13is completely broken effectively. It's
31:16full of OTC transactions, right? And
31:19it's missing a place to settle. And so,
31:21like that that is like the biggest
31:23opportunity right now, and there's a lot
31:25of people running at it. Um, but I think
31:28like the next after that, I I'm really
31:30excited is like this just broader
31:32commerce supply chain, like really doing
31:34kind of business-to-business
31:36relationships, and kind of leveraging
31:38web 3 and AI in one place to really
31:40deliver, like
31:42effectively compressing all of the kind
31:45of
31:46I would say like toxic toxic costs of
31:48doing business.
31:51>> I feel like I'm going to be in an Ilya
31:52shell, but which I it seems to me like
31:54it's something intent
31:58thing. Like, I'm not quite sure what it
31:59is, but as more and more of all of our
32:04interface and work starts to route
32:06through harnesses, whether it's an
32:08intent to get inference or an intent to
32:10get data or an intent to execute a trade
32:12or an intent to do commerce, like it
32:14really is taking a human idea, packaging
32:17it up into some standard format, and
32:20then figuring out the best execution for
32:22it across a bunch of different
32:23providers. And I'm not sure what exactly
32:26the shape of that thing is or how the
32:28market structure of that marketplace
32:31exists. Um, but, you know, I I was just
32:34talking with someone on my team, and
32:35it's like it's intents and order books.
32:37It's like aggregators and marketplaces.
32:39Like you have the place where the supply
32:42builds and then you have the place where
32:44the demand aggregates. And those two
32:46things are going to come together in
32:48some format that I expect will be very
32:51agentic and very crypto and very very
32:54valuable. And so that is the thing that
32:56I think is probably going to emerge in
32:58the next few years.
32:59>> Yeah, there's you know, been a lot of
33:01talk of like outcome-based pricing.
33:04Instead of paying per token, you pay for
33:06what your outcome is and then
33:07And then that that that's sort of like
33:09the AI language for the same thing. So
33:11it's like I think I think this is
33:12coming. It's emerging. So
33:14with that we're out of time. So thanks
33:16everyone. Hope hope you got something
33:18interesting.
33:18>> Thanks everybody. Thank you.

About this transcript

This page contains the full transcript of The Agent Economy: Why AI Needs Crypto - TOKEN2049 Singapore 2026 by TOKEN2049, generated from the public captions YouTube serves with the video. The transcript has 6,278 words across 951 segments, with the original timestamps preserved so you can click any line to jump to that moment in the embedded player.

What you can do with it

Use the transcript to take notes, quote the speaker, build a study guide, generate a summary with ChatGPT or Claude via the YouTube Summary tool, or export it as a timed subtitle file with YouTube to SRT. You can also re-open it in the transcriber to translate the transcript into 100+ languages.

Free YouTube transcript tool

YouTube2Text is a free YouTube transcript generator — no signup, no daily limit. Paste any YouTube link and get the full transcript instantly, with timestamps, click-to-jump, translation to 100+ languages, AI prompts for ChatGPT, Claude, and Gemini, and exports to TXT, SRT, VTT, or Markdown.