← Markets Outlook
Markets Outlook
Why You Should Never Give AI Your Private Key | Markets Outlook
00:18:39
Holonym co-founder and CEO Shady El Damaty on AI agents and crypto’s builder season.
Loading summary
Transcript19 lines
- [00:00]
A
Don't ever give your agent a private key. Don't, don't do it. It's the last thing you want to do. If possible, don't even have a private key. It's probably best to take your key and shard it into pieces.
- [00:19]
B
Hey, everyone, you're watching markets outlook on CoinDesk. I'm Jen Senasi and today I'm joined by Holonim CEO and co founder Shadi Aldamadi. Shadi, how's it going?
- [00:29]
A
It's good. I'm staying optimistic. Somehow in a market structure that is all doom and gloom, I can tell you my secret if you like.
- [00:37]
B
Yes, please do.
- [00:38]
A
Pain is temporary. That's it.
- [00:43]
B
I mean, I think that's a good reminder given what's been going on in the markets. If we, if we look at the volatility, it feels like every time it feels like we're at a bottom where it feels like there's an end in sight, something else happens. So as someone who's just like really focused on the plumbing, who's really focused on the building, how are you watching the volatility? Are you just telling yourself pain is temporary, keeping your head down?
- [01:08]
A
Yeah, absolutely. I think there's only two scenarios currently. One is the doomer scenario that the market structure of the wider economy is reaching this transition point where the assumptions that we have about where economic productivity is coming from, where the most valuable commodities and services really are coming from, and what their future future say, like productivity rates are going to look like are going to change. The other way to think about it is that this is just another leg in the cycle. I'm feeling pretty comfortable that, you know, we're going to, this is a builder season. The next six to nine months, we're going to see some of the most transport transformational applications, we're gonna see some of the most transformational applications in crypto come out, and it's gonna be just like the year and a half before DeFi summer in 2020, where the future of the Internet really was being paved one code line at a time.
- [02:13]
B
If the most transformative apps are gonna come out in the next six to nine months, I would assume we need retail users here. We need people who are interested in using the apps to use them to prove out their use cases and encourage more people to get into the crypto ecosystem. So how are you thinking about that? I mean, is a lack of retail users an issue for folks who are building right now? How are you thinking about acquiring users?
- [02:39]
A
Yeah, it's interesting. I think a Lot of folks when they talk about retail, they're usually talking about, I think they usually mean like a, they're not very specific but they're usually talking about people that trade. Right. So people that are coming in buying tokens and then expecting some huge payoff or some huge roi. Right. Like they got in early on the next bitcoin. Now I think we're going to see less and less of that as the crypto ecosystem matures into becoming a services based economy and less a speculation based economy. I think the meme coin cycle really taught a lot of VCs lessons. A lot of VCs completely lost their funds or post some pretty incredible, let's say ROIs on in the other direction on some, some, some meme coin speculation. And really that was kind of, I think characterized by the Trump family participating in crypto and really sucking out a lot of that liquidity. So I think the scars are still really fresh there. Now this next wave of retail is going to be more, I think consumer based where we're already seeing this, that stablecoin usage is through the, through the roof. I think there's, you know, we've never seen this much adoption of crypto throughout the world being placed since really critical points in the economy. We're seeing traditional finance players begin to actually build out their own defi decentralized finance stocks where they really want to take advantage of the liquidity, the ability to unlock new types of volumes based businesses and volume space trading that wasn't possible before. So yeah, I think it'll look like probably just a boring old financial trading app, a new Robinhood status, a new Robinhood, say application updates or perhaps a new refresh to the Revolut front end that you're used to using. And crypto is going to be powering that and I think these narrative elements are going to build up bit by bit until we reach this head when people are like hey, wait a second. Protocol metrics are through the roof, transaction metrics are through the roof. And that's going to be part of crypto's journey to find product market fit beyond just these speculation use cases. And this is happening already.
- [05:04]
B
Now before we dive deeper into this, I just want to level set, talk to me a little bit about Holonim and what you're building at Holonim, how it's going to I guess address some of the challenges that we've been talking about and what you're planning to launch in the next six to 10, six to nine months.
- [05:20]
A
Yeah, I'd Be delighted to Holonim. Most of most people know Holonim as a ZK identity protocol. We were one of the first projects back in 2021 and 2022 to really demonstrate real world utility for zero knowledge identity. We kind of really cut our teeth off of real world use cases of crypto applied to fundraising. We worked with Andrew Yang's Humanity Forward organization to be able to raise as much funds as possible, not just from like traditional campaign outlets or you know, traditional campaign strategies, but raising directly from from crypto donors. They had a kind of a tough time though, getting their heads around the technology because they weren't able to kyc all of their users and they weren't able to prove that their users were US residents. So that's where we really entered the market and demonstrated IDs utility for things outside of, I think, you know, say, for example, you know, civil resistance and stuff like that. So yeah, we helped them out by being able to prove on chain with full privacy which of their donors were US residents or not. And that was really helpful for them to actually put their funds to use. So a little bit later, we pivoted a little bit more strongly into the wallet infrastructure ecosystem. We saw an opportunity within CKID that if you generate your zero knowledge identity proof and you want to take it with you across devices, you really stuck with the private key that you have. If you're creating it with a metamask, you inherit the security and the portability concerns with that as well. So we started building out what's called a wallet as a protocol. It's like a wallet as a service, but on chain, decentralized, with some stronger security, expect security assumptions and a much more holistic threat model. And yeah, that's kind of where we're focused on on today.
- [07:20]
B
So we were just talking about, you know, the typical retail trader, you know, probably not going to be the group that drives the next cycle. And I, I'm hearing more and more about crypto, maybe many applications in crypto, maybe not being built for humans, maybe they're actually being built for AI agents and AI use cases. Talk to me a little bit about how you're thinking about that. You talk about building a wallet, you talk about private keys. How are you thinking about that future when it comes to human users versus AI users?
- [07:58]
A
Yeah, great question. I have a PhD in neuroscience and I was kind of drilled to be skeptical of any new claims, especially when it comes to technical stuff that involves a lot of base assumptions that are all nested within each other. So ChatGPT was, I think, like a really kind of pivotal moment in where technology is going. And over the past two years we've seen an iteration and improvement of the technology. So like, whether you believe in AGI or not, this LLM technology is here to stay. And it's found a really, particularly, I would say, tight niche within for coding for developers. And no product, AI product really epitomizes that beyond CLAUDE code, I think most developers across tech, whether in crypto or, you know, in different sectors of tech, if you're writing code and you're not using claude, you're effectively kind of handicapping yourself. So what does it mean to use Claude? CLAUDE really shines when you set up what are called agentic workflows. Agentic workflows is kind of, when you take that, say, kind of, you know, basic interaction that you have with an LLM, instead of just casually typing back and forth with it, you set up a series of steps or recipes or what CLAUDE calls skills to accomplish very specific tasks. And one of the most important aspects of an agentic workflow is this self improvement recursive loop. So you ask an agent to pull information from multiple data sources, generate some content or some materials, then deploy those materials to use, say by putting them to market, sourcing BD deals or writing code and then evaluating that output against some objective metric like build errors or customer clicks or deal closures. And this loop can be fully automated. So now what we're seeing are people finally kind of figuring out how to build these pipelines and slowly starting to automate these things, where now an individual will have tens or hundreds of agents simultaneously testing, not just a B testing, but A through Z testing and being able to do things that weren't quite possible before at the rate that we've seen. What does this mean? It means that software is now consuming business operations and bringing business operations closer to the speed of light. And as this is happening, any sort of friction points are being kind of like an immune system, antibodies identifying threats or blockers. It's the same thing with agents in their march for optimization for these business objective functions. You're seeing SaaS friction points being targeted and slowly automated away, I should say friction automated away at a pretty fast rate actually. It kind of feels slow when, you know, day to day at the pace AI moves. But the, the main kind of, I guess, point that I want to make here is agents need to move as fast as possible and then eventually they'll hit the human layer, which is the major roadblock. This involves Things like AML checks, sanctions checks, regulations, kyc, walking into a bank and setting up a bank account. And this is part of the, I think core realization that many in the crypto space have been having is that hey, like crypto never needed to solve the UX problem. Crypto is machine readable, which makes it a perfect endpoint with a couple, you know, slight additions and modifications could be made as the perfect endpoint for agents. So these agents can operate closer to their speed of light limits and be able to unlock new business use cases that we haven't seen before. So everything from spinning up Amazon web stores, to placing orders at automated factories, to scheduling shipments and placing orders for freight across different ports of call and being able to automate away a lot of the regulatory paperwork. So crypto is going to be sitting at the intersection of this because it'll form this source of truth for agents to be able to always point to a ledger, say this is the balance, this is the claim, this is the standard operating procedure for this process. Smart contracts are going to unlock I think a lot of this economy. And the one thing I think you asked about is the private key. So for all of this to occur, agents need to be able to interact on chain and they can't interact on chain without having a private key. So then we get into some really weird kind of situations about what happens when agents own their private key. Is it their coins or is it the system administrator's coins? And how do you navigate that very tricky alignment problem? That's where a lot of our research has been focused over the past six to nine months.
- [13:17]
B
What have you found and what do you think?
- [13:20]
A
Don't ever give your agent a private key. Don't do it. It's the last thing you want to do. If possible, don't even have a private key. It's probably best to take your key and shard it into pieces. So this is a little bit, it's closer to some of the work that's been done in the threshold multi party compute world. Folks like Fireblocks Dynamic and others have really pioneered here. But one really critical technology that's been missing is this ability for two parties to come together and generate a signature. So this has been called the two party ECDSA problem for the past, I would say, I think like, yeah, like past 10 years it's been identified as one of the most difficult, important problems for scaling the wide scale usage of cryptography in decentralized networks. Like how can you get two networks to come together and come to consensus and compute a signature. Turns out that the computational overhead is really, really high unless you can solve some of these core assumptions. And until recently this was solved, this was unsolved. So a team out of Israel by the name of Dwallet Labs was able to crack this and they built out an entire protocol called the EKA network that allows you to do two party ECDSA or any really signature in a decentralized network. So we've taken that primitive and now we've brought it to Agentic payments. So how do we solve the AI alignment problem? The way that we do it is we give the user one share of a key and then we give a network the other share of the key. And an agent, when it receives some sort of, say, delegation or is called upon to perform an act on behalf of the user, the agent has to request the key signature from the network. So the network executes the agent's instructions and does a bunch of guardrail checks that make sure that the agent, the source code hasn't changed, that it's not trying to transfer more than it allowed, than it's been allowed to. It might do some basic security checks as well and can execute any sort of guardrail logic. Once that's verified and executes to true, good to go, then those two key shares can be combined to actually generate the signature. So we think this design pattern in general will become probably really important as we build AI systems that aren't sort of fully, that aren't, I should say like renegade, but rather are fully reliant on their owner or responsible party for guiding and being able to prove their, their behavior.
- [16:10]
B
And Shadi, the last thing I want to ask you, I know that you think a lot about where this is all going, crypto's end game. If, if I will. What does that look like for you when you think about what you're building towards, what all of this is for? What is that?
- [16:27]
A
Yeah, it's interesting. Like, I think if you asked me this question maybe, I don't know, like a month ago I would have said, yeah, we'll probably see something like what we've seen in China with super apps, but kind of like a jungle of them with highly diversified niche super apps that are fully integrated from the L1 level all the way up to the distribution UI level. We've seen examples of this, right? Like with World Chain, World app, World id, Uniswap, Unit App, Unichain, Base App, Base Chain. There's been tons of examples of these fully vertically integrated super apps that have been hyper specializing in one niche or the other. I think while we'll probably still see a move in this direction because whoever owns the user interface is going to own where most value accrual occurs and infrastructure just is kind of like a commodity at that point, I think agents and AI is going to rapidly change just like how we expect value to flow across the Internet. So whether the front end, like as we think about it, when we build front ends we put a lot of time and care into user experience and how we brand things. Well, what happens when you can fully automate and test this in parallel? And we might even see a world where front ends replace ads altogether, where front ends themselves are created almost like on the spot to identify a particular customer segment or a user profile. So yeah, I think what we'll end up seeing is probably some version of the super app thesis, but where maybe in five, six, seven years we'll see a hyper diversification where consumer preferences become very tailored and targeted as development costs go down and accessible information on consumer goes up and we're able to, yeah, I think like create some new economic opportunities that just weren't possible before.
- [18:27]
B
Shani, thank you so much for joining Markets Outlook today. It was a pleasure having you on and we'll be following your journey closely.
- [18:36]
A
Thank you.