Brian Armstrong on agentic finance at Coinbase: trading via AI agents growing 20% week over week

Nov 17, 2026 · Full transcript · This transcript is auto-generated and may contain errors.

Featuring Brian Armstrong

Speaker 3: There he is.

Speaker 1: Brian, great to see you guys.

Speaker 3: Good to see you.

Speaker 1: How you doing?

Speaker 3: Too long.

Speaker 4: Which one should I use this?

Speaker 1: Yeah. Throw that on. The microphone goes on the left side and then it's gonna go right in front of your mouth because it's a little bit loud here but... What do you think?

Speaker 3: What do you think?

Speaker 4: So far so good.

Speaker 3: This thing is... The scale is just is just wild. Yeah.

Speaker 1: It's crazy.

Speaker 3: They've done a great job with it.

Speaker 4: Yeah. Yeah. The massive stadium.

Speaker 1: Yeah. How how are things going in your world? Give us the latest in in Coinbase.

Speaker 4: Yeah. Well, we're really excited about agentic finance Yeah. Right now. You know, crypto has decentralized financial system, very important for freedom. We also wanna decentralized AI Okay. And make sure that there's not a couple just big monolithic ones. You know? It needs to be an an economy of open weight models that are all collaborating together to outperform the frontier. So we're building the infrastructure for that. Accounts, payment rails.

Speaker 1: Yeah.

Speaker 4: I think it's gonna be very good for the future.

Speaker 1: Yeah. Well, I mean, what what does that actually look like? If if someone... I... I've noticed this with the the video models. A lot of the big labs, they don't really have great video models. Some of them do, but it's kind of hodgepodge. And you can always use computer use to go put down your credit card and get an API key. And it's just this extra step, and it feels like it's it's the weirdest thing that it can it can, you know, write all this code, but it can't fill out a form for me. What does that look like in the future?

Speaker 4: Yeah. Well, now the the frontier models are great generalists. Yeah. They can do everything from scientific research to Shakespeare to bedtime stories. What we've seen is that open weight models that are fine tuned to a narrow task. Like it could be compliance or making an ad or like doing voice or like, you know, even just ordering your pizza or whatever. Yeah. Especially if it's trained on proprietary data sets that are internal to companies. It actually outperforms the frontier. Yeah. Then if you get a team of them collab or a swarm, you know, of these agents working together, then it can be really powerful. That... I mean, we see the same thing with human beings. Like, a team of people... Of specialists can outperform a single person no matter how smart they are. And then you start to get companies interfacing with each other. So Hayek is like this economist who famously talked about this. Like, how do you get millions of independent actors in an economy to coordinate? Well, use prices and use markets, and that's what we think the next step is for the agentic economy.

Speaker 1: Yeah. What do you think the overall, tech community can learn from crypto when it comes to security? I mean, the cyber attacks are all over the place. There's systems that are going down every other day. Bitcoin's famously never been hacked. But, like, what else can you tell us about the... What it takes to win as a defender in cyber?

Speaker 4: Yeah. Well, trust is definitely paramount in these moments. And our first reaction is normally to go collaborate with the other folks in the industry to mitigate the issue. But Sure. When I zoom out, I mean, the big thing I think about is that AI is very powerful on defense if you use it correctly. In fact, I think it favors the defender if you use it on defense. So an example is in every line of code that we write at Coinbase before it goes to production gets reviewed by one of these cyber expert models. And so we can look for any kind of issue or vulnerability upfront. Right? I mean, there's other things that we do like, you know, more than 99% of the funds are stored entirely offline and cold storage distributed in places around the world that we've gotten patents on and audited by all of these folks. That's kind of what allows us to be the largest custodian crypto in the world and and benefit in these moments. And then part of it's also, you know, we just got like a national trust charter approval from the Oh, That's OCC. So these are, you know, the regulatory audits that happen give people a strong sense of that security as well.

Speaker 1: Yeah. I know you're also following bio very closely. What are you seeing there? Dario in the Anthropic Roadshow was talking about the potential of curing all diseases, and I think he said a decade. What are you excited about? More interestingly, like, it feels like we would... The the scenario where all diseases get can... Get get cured the same week seems kind of crazy. I feel like maybe peanut allergies go first and then cancer comes later or something. I don't know. Something's gotta be more tractable. But how are you kind of decompiling that industry and trying to understand what's going on there?

Speaker 4: Yeah. Well, so the AI layer is definitely getting much more powerful. But that's all in silico. It's kinda hypothetical. Sure. And to be honest, we don't really fully understand how biology works yet. And so you need to come from the in silico layer where you're testing millions of hypotheses and then actually test them for real in a wet lab. So Yeah. With New Limit, which is the longevity biotech that I co founded, that's exactly what we're doing. So we start with AI at the top to test millions of hypotheses or evaluate them. We take the most promising ones and go through our wet lab in high throughput screens. Yep. We're trying to reprogram human cells to make them restore function that they had when you were younger. We take the most promising ones from those high throughput screens and run them through functional assays and mouse models and things. And then the best of those, take it to human trials. So, first human trials, clinical trials will be happening next year.

Speaker 1: Next year.

Speaker 4: Yeah. And we've gotten some really promising early results, but that that's gonna be the next bottleneck. It's really what do you do with physical atoms moving around? Because we haven't gotten humanoid robots even to successfully be a barista. Sure. So they're not ready to be like in the wet lab in a bio biotech company. Once they solve Barista, I think we'll we'll get the human the humanoid robots in the lab. No.

Speaker 3: I don't know. I think Barista might be a... I think it's a higher art than than the wet lab, personally. I think you might...

Speaker 1: I think

Speaker 3: the wet lab might come first.

Speaker 1: I wanna agree with you, but, like, this year was so weird because, like, hugging face happened before I saw someone book a haircut with AI. And so people are just... Like like, the diffusion, things happen in a really weird order where you'll still get hallucinations and AI can't tell, like, a funny joke, but then it can solve a millennium math prize. But I take your point that like in the wet lab, it is a technical thing and in barista probably comes first.

Speaker 4: Yeah. Yeah, exactly. And of course, the regulatory pathway. I mean, we might see some like, you know, these kind of compounding pharmacies where people are doing their own We GLP might see sort of like n of one type experimentation people do in the biohack community and things like that to sort of get data sooner. But like there's gonna be no substitute to really sell a drug into the broadest markets. You've gotta go through these clinical trials, Yeah. Which still, for good reason, takes quite some time.

Speaker 1: Yeah. I would assume that there's...

Speaker 3: Talk about the potential to, you know, assuming continued breakthroughs, can we get to the point where we can just sip like perfectly simulate human biology, which would of course need human studies to actually verify and things like that. But like the the the full, you know, the thousand x speed up would become... Come from like one to one simulation. Right? Is

Speaker 6: that Yeah.

Speaker 3: Like how how does the industry actually think about that because it's easy for a podcaster to say.

Speaker 4: I think we're still quite a ways off. Right? So Yeah. You know, the DeepMind and these folks kinda got the Nobel Prize for simulating a single protein, like protein folding. Okay. No one has actually gone from single protein to a cell, which is like orders of magnitude more complexity. That's just a single cell. Now your your cells in your body also have communication pathways between them. So to to go from a cell to like a whole organism is another many orders of magnitude. So I think we're probably gonna have to have some like quantum computing breakthrough or something to get this kind of where you could simulate just even a human cell or like a Lab Okay. Bet. Yeah.

Speaker 6: We'll do it. We'll do it.

Speaker 1: Have you been acquisitive this year? Are you thinking about more acquisitions? Is crypto in a consolidation moment? Is that an opportunity? I know you've brought in some great talent in the past, but what's been the thinking this year going forward?

Speaker 4: Yeah. We we... I mean, we acquired about, I think, 10 companies or so last year. Maybe it's a it's a rolling number.

Speaker 5: Sure.

Speaker 4: So we're always looking at different talent pools to bring in, around like agentic finance is the big one coming together. I mean, we have this team, Coinbase for agents is the product. You can connect your agent to your Coinbase account. And a lot of people are doing trading with it. That's growing like 20% week over week right now. Wow. It's crazy. People doing all these interesting trading strategies, but they're also doing it to just solve more mundane tasks like, what's... Figure out my target portfolio allocation and help me save money on taxes and put in a bunch of limit orders like if the the market corrects down and so I just can get better pricing over time. So we're seeing people do that kind of stuff. We're looking at really interest... There's so many agentic finance teams out there. We're meeting a lot of them that could come in and be additive to that.

Speaker 1: So, I mean, we've seen other financial institutions kind of pull back from letting the agents run wild in their system. What what gave you confidence to take that stance?

Speaker 4: Well, that's kind of what we're good at is like we take frontier technologies like crypto AI and we make them trusted and safe. Sure. So yeah, you shouldn't just let an AI come in there like willy nilly. But what we can do is put say, controls and permissions around it to make

Speaker 1: sure It's probably already existed with APIs. Right?

Speaker 4: The APIs already had many, but we're we're making new ones that are tailored to AI agents. So you can segregate funds into a certain account. You could you could give it certain permissions around like what... How much leverage to use or not. Things like that.

Speaker 3: Yeah. The interesting thing to me about this moment for you guys is like because because like cyber has been keeping you guys up at night for your entire history, you're just like actually more prepared for this moment because there's more risk. But at the same time, you're not like a a software company or some cloud or inference platform that's like, wow, we need to care about this. And we didn't really have like... We we haven't been doing this for a decade. It's a very new problem.

Speaker 4: That's right. We always have to remain vigilant on this stuff. But one thing that we did more than a decade ago when we were designing our cold storage system is we said, all right, we have to operate under the assumption that at some point, a developer could get malware on their laptop. And it must be resilient to that. In fact, several developers could have malware on their laptop. And so I think that's a new foundational principle every company has to think about is the key pieces of your infrastructure have to be resilient to that because eventually someone's gonna get tricked.

Speaker 3: Yep. Also the supply... Like software supply chain too that that a lot of companies are figuring out right now. That's something that I think the crypto industry has had to be thinking about as well. Right? It's like Yeah. What data do other companies get and what and what kind of access and that kind of thing.

Speaker 1: Yeah. It's always been higher stakes because it's like, if you put your There's money on here, it's over. Yeah.

Speaker 4: A lot of people... October was just such a bad month for cyber incidents and it's even for companies that didn't have an issue themselves.

Speaker 3: Don't worry. November will be even worse.

Speaker 4: Yeah. But you have to worry about your vendors.

Speaker 3: That's what was saying.

Speaker 4: So, I mean, we... A lot of our vendors, we put them through their paces on these SOC audits and things like that. But I mean, we're actually really trying to whittle down the list of vendors that we use now because some of it you can recreate in house and it just gives more control.

Speaker 3: Yeah. That that to me is like the the part of the new SaaS apocalypse narrative that I've been thinking about is like, you just don't wanna trust other vendors in the same way that you did Yeah. A year ago.

Speaker 1: Because if you have access to some front end model, wanna every line of code, and then then then there's a line that just says

Speaker 3: Just why have it suspended? Yeah.

Speaker 1: I don't know if they're

Speaker 3: Yes.

Speaker 6: It'll take us, like,

Speaker 3: three weeks to build and and more time to maintain. But it's audible. But we own it.

Speaker 1: Yeah. Yeah.

Speaker 4: Yeah. Yeah. There's definitely a flight to quality in these moments.

Speaker 1: Can you take me through the recent history of, Quantum and its relationship to crypto? I know, like, didn't, like, some of the timelines got pulled forward, and then I'm seeing all the math breakthroughs that the models are doing, and I'm thinking maybe things will go faster. But what is the true bottleneck there? What is the risk? How have you been responding to that?

Speaker 4: Yeah. Okay. Well, I mean, one thing is that we put together this Bitcoin Quantum Advisory Council. Or I should say we're a member of it. There's actually Dan Bonnet at Stanford and others have really been hosting it. Sure. And they've had some in person meetings, have been really great with the community just to kind of figure out what the path to upgrade Bitcoin to post quantum cryptography might be. And there's some... The Bitcoin community has these Bitcoin improvement proposals, the BIPs. And there's a good... There's a couple of good ones out there like BIP three sixty that are creating a good path. And so we want to see, as a member of the community, what we can do to contribute to that with engineers or open source contributions. With all the AI research out there on math proofs and things, we haven't seen anything that makes me worried too much about the core cryptography that's in Bitcoin today. But I do think that quantum computers could be... Like, the creation of quantum computers could be accelerated by AI.

Speaker 1: Yep.

Speaker 4: That's something that we have to get ahead of. I don't think it's imminent, the risk, but I think that it's better to be early, you know, than late to something like that. But you

Speaker 1: can't currently vibe code a quantum computer in your garage. Correct. You'd need certain, like, super cool magnets and stuff. Like, you need hard... It's a hard tech project.

Speaker 4: Exactly. So it's gonna take

Speaker 1: it's gonna take us

Speaker 6: a The bunch of thing that

Speaker 3: actually makes me more optimistic on this is that, like, right now, the world is having to contend with, like, complex coordination, specifically around cyber, which is gonna be the same set of capabilities that I think we're gonna need to use in some of these other domains and potentially quantum as well.

Speaker 1: Yeah. For sure.

Speaker 3: And as we're speaking, apparently, Apple is having multiple

Speaker 1: Outages?

Speaker 3: System outages around iCloud, Apple Pay, Business Manager.

Speaker 1: Okay. Well, we'll let you get back to your day. I'm sure it's gonna be Great see you.

Speaker 3: Hopefully, see you around later.

Speaker 4: Thanks for going on. Good to you.

Speaker 1: Still up. The stream is still up, which is remarkable in this day and age. Thank you so much, Brian, for coming on the show. We have, Chase Lochmiller from Crusoe coming on in just a minute. We will see, how he's doing.

Speaker 3: 6.2 is out. Okay. The official OpenAI account says some of you have been asking very politely. Some of you have just been asking. 6.2 Astra and Soul are now available. Excited to see what people do with these new models. Trying to figure out what's happening with this Apple outage.

Speaker 1: User reports, show problems with Apple support. ICloud and Apple Pay down detectors reporting this. Full list. Apple ID login, sign in with Apple, Apple

Speaker 6: Pay, lots of stuff.

Speaker 3: Yeah. The crazy when you start thinking about, okay. What is the economic impact of all this stuff? It's like, cool. Apple Pay going down. Yeah. FAA.

Speaker 1: Yeah. Yeah. Even even one day is like is like billions and billions of dollars of of of economic activity. Truly, truly remarkable. Well, I I hope that they can get everything back up very quickly. Tyler has the full list here, which is kind of everything in the Apple ecosystem. I don't see anything on here that's not an Apple service that I would see. Apple Music, I guess Apple TV is still up. You can still watch the f one movie.

Speaker 3: What about the Vision Pro?

Speaker 1: Vision Pro.

Speaker 3: That's what you're you're frantically like, is the Vision Pro gonna be working today? Thread guy apparently is here at the event and says, Druck just got on stage and said it's over at the Long AI conference. Chad.

Speaker 1: Wow. Insane. Well, we have Chase Lochmiller coming on the show to talk more about everything happening here at TomorrowX. Chase, welcome to the show. How are you doing?