This is Defi Decoded, the conversation show with the leaders, builders, and investors pushing the internet's next economic frontier, hosted by Alex Tapscott. This podcast is for informational and educational purposes only and does not constitute investment advice. Nothing discussed on this show should be considered a recommendation to buy, sell, or hold any securities or digital assets. The hosts and guests may own positions in some of the assets mentioned. Listeners should conduct their own research and consult with qualified financial, legal, and tax professionals before making any investment decisions.
Hello and welcome back to another episode of Defi Decoded. I am Alex Tapscott. Crypto venture capital is evolving. This is no longer a market where money alone is the differentiator. Increasingly, the edge comes from a deeper technical understanding. Whether it's evaluating new blockchains, designing token economics, or navigating and understanding the intersection of AI and crypto, a big topic of this show, the best investors increasingly look like engineers. You don't underwrite what you can't understand. And that's why firms like Electric Capital, who pair capital with real technical fluency, seem to be well positioned for what comes next. And today we're lucky to be joined by Avichal Garg and Ken DeDirk of Electric Capital, which is really one of the more technically grounded, thesis-driven investors in crypto, and they're also repeat guests here at Defi Decoded. So for those who don't know Electric Capital and its principles and founders, Avichal Garg is a serial entrepreneur and former executive at Google and Facebook where he led product for a 400-person team driving billions in revenue. Prior to co-founding Electric, he was an early investor in several leading crypto projects like OpenSea, dYdX, Figma, and Notion. Ken DeDirk is a general partner at Electric Capital and a former engineering product leader with over 15 years of experience spanning system software at VMware and building core products like News Feed at Facebook. More recently, he's been focused on applying AI within venture, developing custom LLM systems to power fund operations. I gather that's at Electric Capital. So to my introduction, investors becoming engineers and actually building stuff. We'll be talking about that as well. Electric has backed some of the most important infrastructure and application layer companies in the space from Kraken to Near Protocol and Magic Eden. At a moment when AI is reshaping software and crypto is reasserting its role as trust-minimized infrastructure, Electric has been really articulating this vision of user-owned technology. They put out a great report earlier this year, their 2026 thesis describing a world where agents, capital, and data all become programmable and sovereign. So, we're going to dig into a lot of topics today. The intersection of agentic AI and crypto, the evolving role of tokens and whether tokens are still interesting as an investment asset class, and where the guys see the next wave of value creation coming in crypto and in the technology world more generally. So Ken and Avichal, welcome back to Defi Decoded.
Okay, I'm not just have you like follow me around and go to meetings and stuff.
Well, I'm not totally sure where to start here, but or who to start with, but Avichal, why don't I kick it to you? So, everyone's super excited about AI agents. And we've said on the show that it's great that there are these tools that can do all this kind of interesting knowledge work, but they don't really have a way to move money or to store money or to do transactions or enter into agreements. But they can use blockchains, they can use stablecoins, they can use smart contracts. Do you see the intersection of crypto and AI as being kind of inevitable?
Short answer is yes. We've sort of long thought that. I think there's the way that it's going to happen and there's the way that crypto people wish it were going to happen, which are kind of different. And I think the way the crypto people wish it were to happen is that every agent has its own wallet and is sovereign and is holding tokens or something. I think in practice, in the near term, I think at some point that does happen, you get sovereign agents, but I think in the near term you are more likely to have essentially command line tools and SDKs, and they could be wallets, but kind of backing that is Stripe and Ramp and some of these sort of incumbents that understand what's happening, and then behind that is actually where the money movements are happening with stables. So whether that's on ETH or Solana, the stable ecosystem really sort of is how that money movement ultimately happens. But kind of that intermediate layer of who has the wallet and the wallet infrastructure and the payment gateways and all that kind of stuff probably looks a little bit more legacy for a little bit. But I do think that there is this emerging category of stuff which we think of, we sort of talked about this internally as user-owned software. Like what does it mean to own your own agent? You kind of see this with OpenClaw for anybody who's played around with that. It's sort of this idea that there's a piece of code that is not sitting on some remote server, that's mine and I control it. It somehow feels different, qualitatively it feels different to have your own code running locally. And so I think that will be a segment and I think that's a very interesting and underexplored segment. And I think that sort of sovereign agent universe could be potentially very interesting. It sort of mirrors, and I think there are a lot of threads we could pull on here, but I think it mirrors what's happening in crypto more broadly. In crypto, you have the institutionalization of crypto, which is like stablecoins and layer 1s and tokenization and Apollo is going to come in and BlackRock's going to come in and build funds and tokenized private credit and move it on chain and all this stuff. I think it's great that's going to happen. It's going to make traditional financial markets very, very efficient and 24/7 and more auditable and more secure, which is great. At the same time, I think what you have and what has always made crypto so interesting from the early days is kind of this non-sovereign store of value third space which is not the US space and is not the Chinese space. It's this US dollar denominated thing that no country controls and no country can kick you out of and no regime can kick you out of. And that as an internet-native financial space I think continues to be very interesting and has a lot of growth potential. And I think that will have its own agent ecosystem. Like I think agents are really first-order citizens on the internet and they're first-order citizens in that financial market where in a lot of ways they're second-order citizens in the traditional fintech and financial markets. And so that is going to happen as well. But I think that will take a little bit longer to play out whereas I think the fintech stuff and local agent stuff for using Stripe and using Ramp is very immediate, you can see that happening over the next 12 months.
Can we just double click on that because I think a lot of people do assume within the crypto world that agents are going to be sovereign executing on chain, using like Web3 toolkit, but you're describing something kind of in between where there's a bunch of companies who see this as big business and are almost like building for agents, right? Where they're creating tools that make it easier for agents to do business, like the companies like Stripe and Ramp and these other gateways and platforms. Why is that? Aren't agents, you know, they're machines, they can read code, why would they need to use somebody else's software stack versus just doing stuff on chain? Like I understand why people need convenience and ease of use and the ability to connect to their bank account, all that stuff is super convenient for a person, but why does an agent need that?
Well, okay, so we should talk about that and then I think I'm sure Ken has very specific thoughts. He's been building a lot of our infrastructure internally around how we're identifying the firm, but I think it's because most business today is conducted either between people or people and businesses or businesses and other businesses. And as legal containers, that is an entity that if you just look at the GDP of the world, if you look at the economy, all of the economy of the world today happens between humans and other humans and businesses and other businesses and businesses and humans. And so you have this legacy infrastructure and it makes sense that you would just make that more efficient and agents can plug into that legacy infrastructure. And then this new sovereign agent universe is really an agent-to-agent universe, right? And that as a model for a market or as an economy, I think one day could be bigger than the people-to-people and the business-to-business economy, but that's starting from zero today. Because what we've done is we've invented a new type of entity, right? So for most of human history all you have is people, right? Or maybe the notion of a family or a tribe, right? But you really had sort of individuals and you could sort of say that's the person I'm doing business with or that's the counterparty of the transaction. And then with the joint stock corporation in like 1648, we kind of invented with the Dutch East India Company a new entity for the first time which was you could put in capital and other people could go take risk with that capital and then you could share the rewards. And around 1850, the US and the UK kind of figured out, well, this is a new thing. You should be able to go to the government and instead of going to the king to get a charter, you should be able to do business, right? And it opens up a whole set of questions like, well, what is the liability, right? Like, if I have a contract dispute, how do we sort that out? What's the case law on that? Or if this entity is implicated in some crime, let's say some extreme situation like somebody gets killed, who's liable at the end of the day? Is it the company is somehow liable? We could take the assets of the company, is it the people behind it? Is it the directors of the corporation? And so we have to sort all that stuff out from 1850 to 2000. It took 150 years to sort all this stuff out and it gets into really gnarly stuff if you think about it, right? Like do corporations have free speech rights? And we've kind of gone back and forth on that over the last 150 years. Like what are they allowed to say and what are they not allowed to say and where can they say and what are the bounds on that? And those bounds are different than what humans as entities can say, right? And we haven't sorted any of those things out for agents yet, right? Like does an agent have free speech rights if it's sovereign? I don't know. It's an interesting question. So until we sort all that stuff out, I think most of the economy is going to be in this people-to-people, business-to-business world and that'll take a decade, I think.
Yeah. Well, I know Ken's jumping at the bit to jump in, but this is something that we wrote up recently because a lot of people are talking about, okay, well, an agent can use a stablecoin or an agent can have a wallet, right? But there's this whole business toolkit that was built for people that can't be used by agents. And the things you're describing are really relevant, like putting aside criminal behavior. What about just legal standing? An agent can't sue in court because it's not a person that doesn't have standing. And a lot of business relies on resolving conflicts, and sometimes you need courts to do that. And so, that's just one example where there isn't dispute resolution software for agents that allows them to settle these kinds of disputes on chain, for example. Maybe there is, maybe that's something you guys are funding and we should talk about that, but that's just one of, you know, it's not just payments and how I store value. It's how do agents co-own something together for example?
Yeah, I think it's much more of a legal and regulatory question than any kind of technology question.
Okay, we'll say more about that.
Yeah, I mean look, the parallel for me is like these DeFi smart contracts, right? Like we're finally getting to a point where maybe the SEC is coming out and saying, okay, these kinds of things are okay and these kinds of things aren't okay. And I think just yesterday they were talking about how are the frontends implicated in this and what are the rules around that, right? Like it took us all this time to figure that out because smart contracts, especially the sort of the admin, like new thing in the world, it doesn't fit cleanly into a human-to-human kind of business or legal rule set, right? And so I think we're just, you know, we're five years behind on the agent side, right? Like what does it mean for this thing that can sort of walk like a duck and talk like a duck but isn't actually a duck, how do we treat it legally? And that's a, you know, I think the pattern that I generally see with these things is that it just takes a while for us to get a sense of how this thing behaves and what the risks around it are. And until you get to that point, until you sort of have a broader understanding of how to think about the thing, it's sort of like a very gray area for a very long time. And a lot of people are trying a lot of different things. And I think we're sort of in probably the early stages of that state now. I would say the agents probably really haven't gotten to enough capability to really be plausible as kind of financial actors until very recently.
Or even as autonomous actors, right? Because ultimately agents are still being controlled by or stewarded by...
Sure. I think we've seen enough, like, we've seen enough things like some of this auto research stuff that's coming, like if you give an agent system a pretty clear goal, it's pretty creative about how it gets there these days and so we're starting to see some of this behavior. I don't think anyone's really tried to apply to, like, tried to turn $1,000 into a million dollars kind of situation. But that's like imminent, right?
Wouldn't it be interesting if you told an agent like turn $1,000 into a million dollars and the most efficient way I found to do that was to say launch a Ponzi scheme and defraud a bunch of investors and then...
Who's liable? I mean, all you said was go turn my thousand into a million, not go launch a Ponzi scheme and defraud people.
Yeah. And then if you did that with a sovereign agent running on a blockchain with a bunch of distributed GPUs, what is that, you know, is everyone liable? So there's a lot of these legal questions that we haven't really contemplated and unfortunately at the beginning also a lot of these things are just purely theoretical and they actually, a lot of these things become, we realize that they're impractical for other reasons that we didn't think of at the beginning and so it's not worth all the machinery of producing laws and regulations to account for cases that are theoretically possible but will never actually happen, right? And so I think DeFi is the same way, right? We've seen all the ways in which these things break and don't break and how people get rugged and don't get rugged and that's really informed a lot of the way that we think about this stuff.
I was going to say something else though. You're asking about why do agents need these tools? Like agents actually want good tools as well. That's sort of one of the things that I've realized. They still look different from the APIs and things we have today. But agents can sort of suffer brain damage in the same way that humans can. They have to use bad tools.
That's why they would use Stripe and Ramp and these...
Yeah. I mean look, you can probably ask them to go under the hood and be like oh go figure out some arcane API and go do these things manually. But the more of that manual work that they have to do, the more tokens they're spending to do this, right? And there is sort of an efficiency frontier. So a lot of building an agent system these days is okay, like what are the actual good tools that this thing needs that you can give it to make it really effective at what you're trying to get it to do.
That's fascinating. Okay, Ken, since we're on the topic of building, in the bio, I noticed that it said that you're now spending a lot of your time building within Electric Capital, trying to make fund operations smoother, building capabilities. How much of your time now do you see yourself spending coding or I guess not coding necessarily but helping to build software versus doing investing. Is this like a new thing or is this something you've always done but you're...
Yeah. So maybe a short backstory is like I've always been, so we kind of joke that I'm the global head of IT at Electric. But I've always sort of had a little bit of a tendency towards wanting to work on tooling. And we've obviously had this kind of theory at Electric around having this capability to build software should be able to put us at an advantage. It turns out there's a lot of internal software that I've wanted to build for the past seven years. And it just, especially when the team is small, it doesn't make sense to have a GP spend all their time building software, right? But it turns out, now that you have Opus and Claude Code and stuff, that equation's kind of really flipped. So I do consider, I tell people I'm on a little bit of a side quest right now. It's not like I'm not talking to founders at all or anything like that, but whatever free time I have now, I'm spending building some of this internal stuff. And the amount of stuff that I can produce with that amount of time is just literally like 10 to 20x of what we could do before. And so that allows a bunch of us, not just me, there's other engineers on the team, to speedrun a bunch of these info pieces that we just would have been impractical to build before and now we can do that.
Well, you said that it was probably not a good use of your time before these tools are available to be spending it all doing this because you've got other responsibilities. But I feel like what you just described is something that a lot of executives in tech companies are talking about doing more. Like Tobi Lütke is the CEO of Shopify or like Zuckerberg. They're all talking about how they're shipping code and they haven't done that in decades because they've been here. Is that like for you, for them? I'm just curious. Do you think that's actually good use of their time to be doing that kind of work or are they just doing it because they love exploring and experimenting with these tools?
I sort of see two things there. Certainly for them to calibrate on what's possible. I think that's extremely valuable obviously because they need to, everyone needs to reset expectations on what...
They need to know when they ask someone on their team to do something if it's reasonable or not. They can only know that by testing it out basically.
Yeah. I think also to the extent that you believe this stuff is going to transform organizations, I do think leadership creating airspace for this kind of activity, because I think one thing we've found even at a smaller organization like us, there is this natural tension among a lot of employees where they have a day job obviously, they're being asked to produce whatever output that they've been doing for however many years. And there's this new thing, right? And the unfortunate thing about the way that AI works is that you can't look at it and be like, oh, this thing can definitely do the thing that I'm trying to do, right? Like you actually have to work with it. You actually almost have to kind of have a developer mindset at the end of the day to be like, can I mold this thing into actually doing the thing that I wanted to do? But it's a big unknown, right? And it might take you hours and hours and hours to really figure that out and learn the skills to do that. And so you're sort of asking people to okay not be productive in the way that they know how to and go do this other thing on the off chance that maybe we can automate that thing in the future. And that's actually a big ask for a lot of people. And so when the CEO comes and says hey I'm willing to do this. I'm willing to put everything aside because I think this is such a big deal and set that tone. I think that's also a big part of it for these guys.
I would add, like I've been hacking on this on the side just on another computer and it's fascinating. I would add a couple things. So one absolutely understanding these things and understanding where the frontier is because it moves so quickly that actually getting your hands dirty changes how you think about it. So like I was playing with the Gemma 4 models which are all local and you kind of get a sense for oh actually this is pretty good. Or you start to get a sense for oh wait on this MacBook M2 with 24 gigs of RAM it just barely doesn't work. And then you intuitively have a sense for where it starts to break. So I think there's actually a lot of value in playing with it because a lot of these founders, the current gen founders, if you look like Patrick Collison, you look at Zuck, you look at Tobi, you look at whoever, everybody's quite young and they all started as product people, they often started as people who had their hands dirty in the product and then over the last decade sort of drifted from that because really everybody was in like a scale game and this resets everything. So I think it's sort of people recalibrating. To Ken's second point, I think is like we don't yet know what the structure of the organization is going to be. And we don't really know if the people who have been successful previously are going to continue to be successful or if it's an entirely different type of person that's going to be successful now, right? And the only way to know that is to sort of play with it and then all of a sudden you build an intuition for like, oh, the critical skill is actually different. Like I can see why this person would be really, really good and this person is no longer good. And now as you retool the organization that's a really, really, really valuable signal. The other thing I think that is going underappreciated with the CEOs building this stuff is there is a long history of people wanting to build tools for themselves because everybody's workflows are weird and unique and different or they have little issues that they need to scratch. But the fixed cost of producing software was so high that it didn't make sense to do that. Like you're going to hire, remember there's a time maybe like 2010ish or something where like super rich people were hiring people to build personal apps for them on their phone because they just had weird, you know, so some billionaire would hire somebody and pay them half a million dollars to build a phone app. Well you can just do that with Claude Code now, right? You can just have somebody spin up an app for you that taps into your workflows and so you've taken the fixed cost to close to zero to doing that and it doesn't need to be production grade, it's just like let me spin up a thing that does an XYZ little thing that I do all the time, have a thing that's like checking land prices in a very weird way that I like to do on the side and it just saves me two hours a week. And I see a lot of that, especially from the CEOs, because a lot of CEOs have weird sort of trade-offs that they made in life to be able to run the business the way that they do. And now you can just spin up software. It's actually faster than trying to convince somebody else to do it for you or waiting for the market to do it. It's just faster and cheaper to just do it yourself.
