Making Onchain Agents Useful | Gordon Liao - Circle

about making onchain agents useful and then we will hear from Jake Buckman from coin fund.

All right, thanks so much for having me here. Uh maybe I'll bring uh some centralization uh to this uh decentralized AI uh set of sessions. Uh so my name is Gordon Leo uh representing circle and today I'll make a presentation um uh how do you how do we actually make onchain agents useful and uh talking about the trust infrastructure of agendic economy so I think the agendic economy has uh arrived or is arriving and I would I would say that u overall I don't think capability is the bottleneck rather the limitation is trust and coming from circle I have uh quite a bit uh that I learned over the years about how to build trust and how to maintain trust and I think there's uh quite a bit that the agent economy could also gain on trust. Uh I do think there is a trust problem today. Uh so earlier uh Eric presented on x42 it's been tremendous progress uh 24 million process in the first seven month.

uh the uh EIP uh 804 that was launched last month already have 35,000 agents registered uh which is amazing uh take up rate uh but overall I would say the economic accountability of these agents are literally zero dollars that is uh you know even if you register for the reputation and identity uh it's not really worth that much you can't go after them uh if they rug Uh so as a brief uh review of where we are today uh XFO2 as mentioned earlier uh it's about the settlement rail for agents. So agent A can pay agent B or pay for API service uh for perhaps a fractional scent. This is coordination that's needed for agenda payment but it doesn't really help the agents to know whether they should pay for something whether that something is worthwhile. where 804 which is a uh newer proposal uh stepped in is registries with three components. Uh one is identity having a ERC uh 721 an FT attached to agent profile and uh two is reputation having ability for other agents to provide feedback uh perhaps aggregating scores and three is for validation uh for third party checks maybe through tees and then economic proofs.

This all sounds great. Uh but at the end of the day, I think a lot of these are uh still problematic. Uh why do I say it's problematic? I think there's a certain limitation to how much reputation identity you can build on chain when there is no real world consequences. um you know I might trust the agent to help me post some stock API and get some stock prices or weather uh price weather signals but it's hard for me as a individual or perhaps an agent to trust another agent with perhaps managing large amount of money even on D5 protocols.

So on the spectrum of uh test value, if you have low test value like API calls, uh reputation might be enough. But if you actually have very high values, uh things that uh could be very problematic if it goes wrong, I would say the economic security which is amount of um locked asset perhaps that could be lost. It's it's actually required. you had to have uh necessity of economic security that's beyond identity and reputation. uh this is because you know you have failures with onchain system as we've seen over the years and this is you know I don't know how many years have been at emer and every year you know you have uh quite a bit of hype but at the end of the day you have a lot of civil uh in this industry uh in in bad activities in tokens uh you have adverse selection the bad agents are motivated uh to fake reputation this is no different than uh online reviews but when you add agents review signals, the signal to noise ratio drops even further.

And lastly, you have no skin in the game, right? The agents are uh purporting to be reputable and really the most they can lose is their current reputation, which is very different from human individual. The most you can lose is you go to jail. This is why uh if you think about I come from a trif background, right? If a asset manager uh misled his investors and uh misallocate, the worst thing that happens, the asset manager actually goes to jail.

Uh like some of the folks uh in this industry in the past. Uh so I say that reputation is not enough. You actually need to have some recourse and the recourse has to be linked to identity and has to be linked to uh perhaps economic security, economic stake. So this morning I pulled up uh the AO4 uh registry and there's 35,000 agents registered. If I rank it by the most number of stars or or the scores, you kind of see there's a bunch of these uh agents.

You know, I click into some of them, a lot of them are kind of like, okay, there's a token associated with it. What does it do? Right? I don't think this is the, you know, this is not the end goal yet. Uh hopefully the agents are more useful later on.

Hopefully they actually do things. Uh but I think this is where the the the shortcomings are currently. This is where uh we are currently and it's we're not at the state of making these agents really useful for economic activities on chain and I circle we're building the economic OS um and which I'll talk about at the very end of the talk. So what do I think uh is needed to build trust and accountability? I think it's probably three pillars.

I briefly touched on it earlier. One is developer linked identity which tells you who built this. It provides real world accountability. Somebody might go to jail like if somebody messed up. Uh two is uh slashable agent staking.

Uh so what is I stake? What what what found is actually uh I stake if a agent perform something malicious. And uh three I would say there you know we have invented quite a bit of um economic security uh ways of uh securing blockchains. A lot of it could be applied to the agentic space. Uh so can we actually have a market price um of agentic security agentic trust and that would generate a continuous trust signal and that is through commonly what we have uh learned over time uh through a token process.

So in the first pillar uh links to agent links linking the agents to the builders. I think again this could be tiered right you could still have anonymous uh agents uh that are for low value activities. Maybe you have uh pseudonomy for linking into a ENS address of a uh of a dev uh if it needs to establish some level of uh confidence but it's not high value. And then if you're going to manage finance or have large values that's dependent on these agents, I think the the agents need to be identified uh traceable back to the developers or at least to an agent and then to an agent back to the developer. Um so I think there's a lot of theory behind this.

Uh you know I'm an economist by training. So uh there's a concept of costly signaling. What is it? It's uh uh making sure that using very costly signals to separate a pooling equilibrium into separating equilibrium. What do I mean by that?

You know, usually a pooling equilibrium just a bunch of agents, a bunch of dabs, you don't know which ones are good, which ones are bad, but if you impose certain amount of cost, now this might not be democratizing, but if you impose a certain amount of cost, you can actually separate the legitimate agents from the malicious agents. uh if you set the cost correctly. Um costly signaling happen in real life all the time. This is why people go to college and get a very fancy degree. A lot of the time is not necessarily the learning but rather it's a signaling.

You could do this pretty easily in uh you know in in JSON files and you could actually create a portfolio for the devs. So it's it's kind of like a restaurant chain right? If one gym one agent behaves badly, it hurts the entire portfolio of the dab or or the agent that's deploying other agents. Uh two is having skinned the game. Uh what that means is having funds stake ETH or USDC or agent tokens that could be slashed.

Um that could perhaps also be funded through continuous payments uh that the agents receive for their services over time. Uh so it could expand and grow along with the services along with the reputation and credibility is established based on how much is actually slashed. I'll skip this. It's just you can automate a slashing or you could go through various process of optimistic uh fraud proof things we have known for a while in in the blockchain space. Uh you could fund the staking through revenue as I mentioned briefly.

Uh so the veteran agents would have more reputation over time as more and more of their cash flow gets uh funding uh gets gets uh staked into a bond and there's a market segmentation effect right clients could select uh or or developers could select how much of uh the cash flow go towards the agents. Lastly uh but not least we have a very good mechanism for uh launching uh ERC20s currently. Um how can we tie this to market price I would say that it's a trust coordination mechanism uh no m coins please in this space and uh market price I think is very effective at um aggregating disperse information better than centralized systems. So putting all this together you can identify who is the agent, what is at stake and uh what is the market value uh a certain agent and putting out all together I think we have all the building pieces. Uh very lastly there's a lot of economic literature and work that's related to this.

I'm not going to go through it. Feel free to follow me if you have interest in learning econ. and uh a obligatory uh mention that we're building ARC uh a internet operating system uh economic operating system uh this is where some of the aging activities will happen and as well as uh on many other chains. Thank you so much.