Fireside Chat with Kartik Talwar & Vitalik Buterin at Pragma Prague 2025

[Music] I know. All right. Well, the one that you've all been waiting for. Let's give a massive round of applause to Vitalik. Wait, do I have a mic or not?

or you're maybe bring me one for convenience. Oh, it works. Okay, perfect. Okay, wait a second. Okay, now Oh, right, right, right.

This thing. Yeah, no, that's good. Okay. Um, this is great. Now the the the playing field is more leveled from uh from my Yes.

Yeah. So, we want to talk about stuff around Dows and I think this is uh how far we got in planning this. So, uh yeah. So Cardik, I guess maybe to start off like what are I mean you have this like very unique and privileged position of like being right at the head of this uh lovely group that can that continues to attract you know amazing developers amazing speakers builders from uh all parts of the Ethereum world. So I'm curious uh like both positive and negative what kinds of things have you seen from the world of PE of existing Dows and people trying to build Dows for in the last couple of years?

Um I'll split this into two sides. One is when we as an organization are working with Dows to get them involved at the events and then what we see from uh from the hackers and then um there's like a separate perspective on actually you know what and each of those have also two threats. So when we're working with organizations, um the part that we've seen kind of that's been a very consistent theme is that it's unanimously extremely hard to work with Dows if you are an organization. M um and those are challenges around not knowing who the directly kind of responsible kind of DRRi is for for for a given task or or a given um outcome to if they want to get involved. Uh we need to figure out how do we accept money from them and do normal compliance which uh requires us to follow the laws from our end and make sure that there's a legal sign, they have the authority, they they have a legal entity or something.

um the the common kind of blockers you run into when you work with a company as a as a company. I work with the DAO as a company. Um on the hacker side, it's more interesting because we kind of just stopped seeing experimentation about two years ago. Uh I remember specifically a massive surge from 2019 to 2021 and um we kind of saw way too many clinical experiments and largely it coincided with a lot of the DeFi revolution. Mhm.

Um cuz people just kind of saw the just the more dows kind of came to be and they started popping up and we saw small tweaks and variations. Um and that also even led us to do a whole hackathon just focused on Dows to see what people can uh kind of come up with or push the limits. But beyond that, they just kind of fell fell flat. So I have not seen enough hacks that would let me say that there's a substantial attempt to play with Dows anymore. Um I have not tried to answer why um or figure out why, but uh my my reasons my guess would be that it's just not as exciting and um and in the context of a hackathon where you have a time constraint, it's probably a negative thing for you to uh build something that requires coordination to show the first example and uh and the value of it.

And um faking that does not make a good demo because you don't get to land the point. Um so actually just like revisiting a bit the the thing that you started with right challenges of working with Dows as an organization. I wonder if uh so like the part maybe not the parts that have to do with compliance because like that feels like understandable. Exactly. It's it's an equal problem no matter the type of DAW you have, but the problems that have to do with uh like it not being an easy thing to negotiate with like is is there any more detail that you can share?

Like obviously okay not to name names. Um we we'll see uh so there's there's a couple of things. one is that um often times you have yeah I'm getting do not share names from uh some people on our team which I think is a smart idea. Um the the part that we've seen as a very common threat is that there's only a single interface to to interacting with a DAO, which means most of them will say either fill out this form if you're in this one of these specific tracks, but like it's a grant or or request to be an event sponsor. Um or go to our forum and then here is a couple posts from the past.

We've seen others do it. use that as a reference and kind of figure out what you need to communicate and then we'll let anybody kind of comment and and drive the conversation. U the problem with that is it is in my opinion a a anti antagonistic default um that you kind of end up proposing which is you get treated like you have to defend from the get- go. um which in some cases is fine but in a lot of cases like it's it should be discovery and it should be around understanding if this is something that's interesting before you move on to the stage where you have to convince one or the other party. Uh but the second you have these public discussions as a default and anybody is allowed to comment on it, it kind of puts you as a defense and you have no idea what you should be defending against without also knowing does the thing I'm trying to convey to the other side will it even reach the right person that should actually be reading this.

So you end up all with all these side quests of fighting people for things or just debating like just minutia like just doesn't irrelevant stuff before you even get to the right conclusion. Um and that ends up just kind of equating to time wasted without any outcome. Okay. No, that may actually make sense that uh I mean I I have uh one story that I can't remember from a couple of years ago. So remember back in 2021 when the uh the dog coins were a big thing, right?

And then remember when the Gitcoin Dow got like 40% of the Akita tokens and then uh the Gitcoin Dow was like basically immediately thrown into this uh negotiation process effectively like between them and some people from the Akita community who uh and and the like basically the fact that uh like they were Adele like ended up putting them at a really big disadvantage, right? Because Adele in an adversarial negotiation process like basically if the internal wires of your brain are open source then uh like effectively you know how to play the game. Yeah. Exactly. The other side knows you know exactly how to give you like how to create the offer that leaves you with you know like epsilon surplus and leaves them with like 100%.

Right. So it's uh so I think in general like this lines up with a lot of the things that I've uh heard right and and I mean things that I've seen right which is basically that mean like one Dows are kind of at a standstill and two is there's a couple of like very specific reasons why and the way that I would give the diagnosis is basically that it feels to me like from like the Dow base kind of settled in this equilibrium where like essentially nobody gets fired for forking compound and uh the problem like basically once you have an established path then like that's when imagination stops that's when progress stops and uh the problem with the DA is that uh if you have like a low competence form of governance then like you don't you can't really tell or like you can't even form consensus around that fact until like it's basically almost too late, Agreed. And I think the the part that I feel like I find slightly annoying is that it's not even the nobody got fired for for working compound as a good argument because you don't know if that is the the maximum whether it's local or global. Um and we just kind of settled on hey just because this was working and it wasn't broken enough it's easy to repeat it. That that should not have been in the norm.

Yeah. I mean one reason why it was I think is like that we don't want to or it's important not to underrate is sort of the Dow as compliance aspect right like we basically had this like very weird legal regime where like number one making an honest investment is illegal but making a thing that you explicitly say in the terms of service this is a joke memecoin is legal. Um, and so basically people like even well-meaning people ended up like repackaging what they wanted to be honest investments into joke memecoins. And so effectively Gary Gendler like massively decreased a whole bunch of people's investor protections. Um and uh number two we uh had like whenever something is used for compliance like that itself creates conformity pressures right because the point of of compliance is you know you're a um cover your um a and uh you know when you're covering it sets the floor.

Yes indeed. And when you cover the easiest way to cover your A is to like basically yeah like points to some other source of legitimacy and copy them right because you know if you're doing something new then like effectively like the you know if the if run comes you know the IF run will actually end up being on you as opposed to if you copy someone else it's like hey like that's you copied them they copied them you get penalized for going out of your way to do something and making it better instead of copying the standard right and like the thing that feels like is happening now is like uh now that the need to do decentralization for regulatory reasons is like reduced by by a lot like I am noticing that there's a lot of projects that are just like giving up on a on uh operational decentralization uh what does operational decentralization um in the sense of uh like basically Dows controlling uh like aspects of uh like protocol development funding um like other things as opposed to just the foundation doing it. No, I agree and and I think um it tracks really well on the the private kind of venture side too which is um we're seeing more and more people actually incorporate companies and set a foundations company in the US which is a a pretty big move because uh you are choosing to uh in a way go back to a decision that you thought you would never make in the last 10 years. Yeah. Okay.

So this is uh why uh you know like why Dows suck. So the next thing I want to talk about is why Dows are necessary. Um um I'd love to do that but just want to understand one kind of point from the past like the last few minutes. I feel like everything we talked about ended up being about process and nothing about the value or the nature. Um are we in agreement on that or do you think there is some nuance there?

What do you mean by value? Uh sorry as in everything we talked about as a pitfall is purely on the fact that um if you were to replace a DAO with a a poorly run company or a group project or a telegram group chat uh none of the answers change. They're they're fully detached from being a DAO in any form. Um right. So I think we want to like decouple like organizational structure from like I guess kind of like legal form.

Right. And uh or maybe like I'm not sure if legal for is the right word. Right. Yeah. I guess I'm just saying like everything that we talked about is a a human process flaw, not anything related to a Dow or being on chain specifically, right?

Okay. Maybe kind of triangulate a bit, right? Like you want to know like find sort of the opposite, you know, like quadrants of the square. Like imagine hypothetically someone made an onchain DAO and the way that that Dow worked was like much more conventional. Like imagine for example every 6 months uh the like token holders elect a CEO and then that CEO has the rights to just approve everything unilaterally like say up to some dollar limit.

How like how what would you predict your experience negotiating a sponsorship with that would look like at least faster than usual. Mhm. Um just because it by default assumes a hierarchical flow of how anywhere from people to initiatives and priorities are organized. Um so you at least get an answer back faster because there's a clear incentive and and a goal that this entity has and there's like just much harder or much stronger accountability. Mhm.

Which I think is the key factor here. Right. And then on the flip side, imagine you were negotiating with a regular company, you know, an LLC or GmbH or whatever, but that company just like happens to have rules that required board approval for literally everything and the board had like 73 members. Um, how would your experience negotiating a sponsorship with that look like? Just the way Dows are right now.

Okay. Yeah. So, I think but but but they can still move. Yeah. Actually, if if it's a board approval with that size, Yes.

Okay. Yeah. So, so I think uh like a lot of the difficult the the challenges like do have to do with like the actual structure, right? Like uh more about sort of how the wires are made rather than whether the or how the wires are organized rather than whether the metal the wires are designed or are created out of is sort of Ethereum smart contract metal or legal enforcement metal. Yeah, I agreed.

Okay. Yeah. So no, I think it's uh like also like an important thing. So okay, so maybe you know why Dows are necessary, right? So um like what are the the biggest users of Dows right now?

Uh DeFi projects and NFT projects. Okay. Um and so the one I would have mentioned is ENS, but but that is an NFT project. Uh so technically ENS is an NFT, but yeah. Yeah.

Yeah. Yeah. Yeah. your C721. Um so okay um so I think uh for D5 project right for a D5 project uh to work like if okay if it's unis swap unis swap can like be yeah like this actually doesn't apply but if it's like say you know like something like a or something like a collateralized stable coin it relies on an oracle right yes the thing is only safe if the oracle is safe if you're But if the oracle is safe.

I see. Yeah, that's fair. Yeah. So like if hypothetically, you know, you had a big CDP for $10 million and then someone broke the oracle, your $10 million will be gone. Okay.

Now, so therefore, for DeFi to work at all, the oracle has to be managed by something. Um, now if the oracle was managed by like one person, then uh like would it be reasonable to call the thing DeFi? You mean the DAO a DeFi Dow? Yeah. Well, like like the or like the oracle, right?

Either it's managed by person or it's managed by multiG or a DAO or something else. That's a tough one because I think you can argue both sides. One on just that the you can say yes based off of the fact that it's it's classified as D5 because of the utility it's providing. And then one is on the tooling aspect of it. you need these five things to be true for it to be considered that.

So, um I feel like you're taking the latter stance in that case. No. But, uh if if there's a price oracle and it's run by one person or one computer, but it still serves the job of fighting you the price, uh it may still count. It may not be as right. What's the thing that you would feel more comfortable putting $10 million in?

uh better asurances on uh having anywhere from fallback to uh uh in a way surge protection, storage protection, a search or just like any any volatility or just um having having some sort of backup plan for how do you actually handle inconsistencies in anywhere from data to stronger consistency on the expectation of liveless, right? Okay. Um and then for yeah I mean for ENS it's uh like a a similar thing, right? where like basically you know like protocol upgrades have to be managed somehow and uh either it's managed by centralized thing or it's managed by decentralized thing right so like I think one thing that's uh interesting for me about the difference between like these kinds of things right and like you're basically trying to motivate like use cases where Dows actually make a lot of sense from some of the earlier use cases is that there's a difference between like organ like things that have to make strategic decisions versus things that have to fulfill like basically some dumb pipe function that's like that's still subjective right so I think one of the arguments that I would make actually is like we should actually view those two very very differently right I agree yeah like if you have an oracle then uh if you had uh a board of 73 people and there was no CEO and each board member just by themselves had to figure out how to maintain the ETHUSD price and then you like took a median of all of them then like that would actually work better than any one of them but but on the other hand like you don't want to negotiate with that. Yes.

Yes. Okay. So like I mean this is uh like basically like there look this is also what I call the you know like the concave versus convex regime right basically where like if you think of the decision space that a thing has to make like basically is a compromise better than a coin flip right and there are situations where a coin flip is better right like uh you know you definitely don't want ETH proc to be hosted like halfway in between here and Berlin um you definitely don't want it hosted halfway in between here in New York. I'm pretty sure that's in the middle of the Atlantic Ocean. ETHUSD values is like probably more accurate than either one of them.

So basically, I think uh one of the arguments is like there is a type of DAO whose job it is to like basically just uh like perform some subjective function where it's like not actually trying to be a live player. Um, yes. The I think the interesting part to me is that while I agree with what you're saying, I feel like you're picking an example that is way less common in our space. The majority of how Dows work right now are purely fund managers. Mh.

And right, they don't need any external data on at least a contract level to to behave. they just need to coordinate sending money from A to B. Agreed. So like basically Yeah. So yeah, so fund allocation is also like one of those you like I think if I like if I had to kind of make a like very high level typology of the Dow space I'd say like things like data oracle like things I mean possibly throw in type court like things um in one bucket um like like data providers is one another one is fund allocators and then a third would be like basically things that have to steer a ship What's an example of the third?

Um something that has I mean like if hyper would ENS be the third one here or the first? I I think ENS has to be the third one. um like so what's okay maybe there's like a there are more popular first one because I guess I'm looking for like what are what's interesting about popular here right is that the the data provider use case like we don't think about it but it like underpins the security of like billions of freaking dollars right like absolutely I guess would you say the maker oracle is the first one here or is it the third one um maker maker oracle is the first one but the the dowo like the maker dowo in the sense of the thing that's deciding to rename itself to sky and like doing all that stuff. Go back to maker. Yes.

Right. Exactly. Like there you don't want compromises, right? Imagine if like we made a compromise between maker and sky and the thing got renamed to my protocol. So the um so there's like a bunch of different spaces, right?

Um and uh like I think these spaces like in some of them you clearly need Dows and in other ones like there is a stronger case that like for example something like the parts of Maker that decide on you know like whether or not to do a rebrand like probably should look more like a corporation. Yeah, I agree. Yeah. Yeah. Okay.

So that's um one one of these kind of things, right? So then like for me the interesting question is uh you know like if we wants to make uh like kind of remake these things from scratch right then uh you know like what kinds of principles would we remake them on top of and like so we talked about coin like basically the existing you know like compound model right and one of the things that struck me about what you said it also confirms what a lot of other people have told me is that like actually a big part of the problem comes from transparency right basically a big part of the problem comes from the fact that there's this public discourse pro process and uh you know if you want a sponsorship you have to like publicly negotiate with a crowd right and uh I mean the whole like the Gitcoin Akita situation thing was actually pretty similar right and like actually you know a negotiation where you're like every single thing that every person says is like simultaneously also public signaling is like like there it creates a lot of problems well that but but it's Yeah. Yes. But I think the the more concrete point there is that you are now suddenly in a conversation with people who have nothing to do with that decision and are providing input that gets treated with the same weight. Okay.

So not enough specialization basically. Uh not even specialization. It's just that like it it's like let's say let's get dinner today, right? and like where where should we go? And and everybody from the audience starts to recommend their preferences for what they think they're feeling like eating tonight.

Uh that's a weird situation uh because they can give input, but you feel like now you have to consider it. Um or if you say, "Oh, I feel like eating, I don't know, Japanese food today." And somebody says loudly, "Oh, no, that sucks." like it creates a group setting that should not be a setting to begin with. Um and that's what negotiating in a doubt looks like, right?

Okay. So, what I guess the dinner for the dinner example, right? Like uh one way you might plausibly impro um improve on this is uh whenever it's obvious that a group is going to make have like needs to have dinners like let's say the subset of the ETH global world that's coming to Prague you make a subdow and then only the subdow makes decisions. Yeah and that's how people were saying so you now have like oh if you want to have a request for a sponsorship or for us to be at an event go post here. So there there is specialization but I think by but by but but the the expectation most Dows and forums set is that anybody is able to uh chime in and say whatever they feel like even if they have zero involvement in the community or are even a token holder in any Yeah.

And so I think one of the questions that like I think is interesting to put to the world, right, and this is uh like part of my goal of trying to reopen the Overton window here is what what if you have Dows that don't have a social norm that just about anybody is able to contribute to anything. So what's the definition there of of a DAO? Just an execution engine of a decision transferring money or something else? I mean at the top level, yes. Um but uh I think that the question is like what is the underlying value that the decentralization layer is providing right [Music] I guess I'm trying to think about how often u are Dows making a decision on the potential impact of a request or a decision versus a more deterministic outcome because I think what you said works really well in those scenarios.

But if I have to say, oh, if I were to do this now, if I were to bet long-term on X, um, I think those are less modelable, so what, you know, what kind of scenarios are less modelable? Let me see if I can I'm trying to figure out what is a good way to communicate that. Um, somebody willing to make a bet on saying, hey, if we have the option to Okay, you actually is a good example. So, a DAO has a mandate to get more people to build on the protocol. Um, very very abstract.

And there's two requests that come in um that says you should uh you should build um or you you should uh hey, we're hosting an accelerator in San Francisco. We're going to get the best engineers and we think uh you giving us $100,000 will get you at least three companies uh three three projects built trying to build companies on you which is like just very very random. And then you see another proposal saying hey we can do the exact same thing but in some small town in some rural area in in also America and we'll try to offer you the same outcome but for $20,000. Right. Um, mathematically, you're like, "Okay, well, if the outcome is three companies, maybe I should get the the cheaper one."

But, yeah, if you are betting on a long-term tech company, maybe it's a higher odds bet that you should spend a h 100,000 because the the odds of that company succeeding out of San Francisco might be higher. Uh so you have some of these types of situations where you want the ability to override a decision because it comes from someone's anywhere from experience to um kind of taking a bet largely and and those are not deterministic outcomes, right? Okay. So like you want to have uh like you want dows to be friendly to the idea of a person making a bet and like sometimes that bet is is going to be non-consensus. Yes.

But but but it's not social because we're talking about removing the social aspect of it which which is not the same as removing a human from endorsing what action should be taken. Agree. Mhm. Right. And so possibly like reducing the amount of communication or maybe like structuring it so it's so you don't like the default is not all to all might actually enable um like more kind of radical bets to be made and that's how usually most companies function.

Indeed. Okay. Now the thing that you'd want is people making radical bets where you might that that might turn out to be a success to just suddenly you know motivates like a bigger change in strategy and then suddenly you global is five times more effective. The thing you don't want is um someone saying like, "Hey, how about we put $100,000 so that my brother can put a te can buy a Tesla and uh you know like here I'm going to give five paragraphs about how he's going to put the word global on on his Tesla and that's going to be marketing for you guys hypothetically speaking. Okay.

Uh so that's um and so that like what that implies right is that there is value in some kind of account. I mean, we've seen Dows put their logos on F1 cars, so it's not even a fake example. Yeah, indeed. Um, so you do want some kind of accountability layer. Um, and so what's interesting there actually is that making the accountability layer decentralized might actually be fine, might actually be best.

Okay, I'm not sure I agree with that. Can you can you help me understand? Um, so let's imagine you have a person who starts a project and uh let's say this project could be starting a new $100,000 accelerator and the project could be yeah put like putting the word ETH global on his brother's Tesla and buying the Tesla. Um then six months later you'd want to have some process that evaluates it. And then if the evaluation turns out negative then um you know presumably you would want to like basically yeah not give that like that person any any execution authority anymore.

And if it's positive then like you might actually want to give them more authority. Um, how is that any different than just having that same person be valued in a shorter time scale if they were still responsible? Like, well, what do you mean validation? Like, cuz like an accelerator like it does take six months to figure out if the company's even No, no, sorry. Just so I understand, did you was your initial premise that you're saying anytime a decision is made, it's going to a new person and then we evaluate if that person should be allowed again?

Not necessarily. I'm saying uh whatever mechanism you have for allocating decisions to people, it could be random, it could be one CEO, it could be whatever then um you know it might be valuable to have a layer 6 months later that basically says like was this actually a good idea? So, so let's say in this case case like that accelerator is one decision this person makes, right? And then the next week they make 10 other decisions, right? So you might you might want to merge and you might want to like evaluate a collection of decisions in totality.

That's fine. Okay. So I think but sorry so your argument is that if someone else usually at random is evaluating this in the future and they determine that this was not a good decision this person should be penalized and not whatever. Well I mean I guess my mom was saying like one is uh like okay do you agree that having some kind of like evaluation mechanism is can be valuable. Yeah absolutely that's how companies work.

Yes. Right. And then two is uh like that evaluation mechanism being decentralized uh in some form is fine. But does that equate to a new jury determining this or something else? What does it mean to have that be decentralized?

Right. It could be a new jury. Um it uh could I mean my like if if you had to ask me for an answer, I would actually say it should be a jury that's a combination of like relative insiders and relative outsiders. Um it I mean in principle could be token holders but I think like I would only support you want some bias. Yeah.

Like basically I would only support like a full outsider jury for decisions whose scope is kind of grand enough that like the entire community would be in a position to basically be insiders. Um no I mean I think that makes sense but I still don't see why you need any new forms of setting that as a norm. You can have like if you're already agreeing to having an hierarchy and having somebody responsible for a decision, right? Uh presumably you would also set up a feedback loop for that person which doesn't have to be outsiders, but it could still be part of a yeah it could be it should be people inside the community. Um so okay like basically you know if you have a mechanism that is um centralized then like either that's that centralized person is reporting to something else that's centralized or they're the king and if they're the king you need a way to throw out a mad king.

Yep. Okay. And so basically Yeah. And so the way to throw out a Mad King like so I would actually So my case for making it decentralized is basically if it's centralized then it's just going to collude with the king and you have a capture risk. Um but how do you ensure that there's no collusion when the jury is evaluating it?

Okay. So this is uh where I think like privacy by default can actually help. So, but we're all just relying on votes. But, but doesn't that jewelry need consensus in terms of like they need to talk to each other instead of pure binary kind of decisions that they make and the aggregate is determining someone's fate? So, it depends right because uh okay so for example are you familiar with how the conclave works like the mechanism for choosing the pope?

Yes. Yeah. Okay. So you know you have the cardinals right and then the cardinals get put in a room and then you know three times a day they make a vote and uh then and between the vote they are allowed to talk to each other but they're also not allowed to talk to the outside world right and so what's interesting to me about that is like it's imposing a communication structure other than full transparency and it's doing that in a very in in an opinionated way with the goal of uh like basically trying you know like maximize quality and trying to minimize I mean capture risk is a big part of it right um so like there have been a lot of uh like historical situations where like lots of like very powerful countries would care who the pope is right okay so like I would so I would argue that the one of the big value propositions of decentralization is capture prevention I agree with that yeah okay so maybe right and So one of the things that makes capture easier is uh like basic one is a smaller number of people and two is uh like much wider communication sorry is the thing we just talked about considered much wider communication or more accountability right so in order to capture someone you have to be able to communicate with yes okay sure okay and so there's also two types of capture one type of capture is what I would call internal capture And internal capture would be when basically the like the people that are meant to be trustees of a thing become more loyal to each other than they become to the thing. External capture is when the trustees to the of the thing all get bribed by the French government.

I know we I just have never picked on France before. I guess we won't be seeing you at ETC this this fall. Uh okay. Okay. Yeah.

Okay. So if you want to avoid external capture then you do the conclave thing. Just not have a private chat I guess. Exactly. Right now, if you want to avoid internal capture, then like you could actually try to like put some walls inside of the conclave, right?

And like you could do lots of things, right? Like you could just uh like basically Yeah. You know, you could put like every cardinal in a separate cell or you could put them in groups of three and let them talk to each other in groups of three. Or you could even put them in separate cells, but let them go on the um go on the internet, but don't tell them who the other cardinals are. Yeah.

Okay, I just want to point out that I think in this example and this design, you're assuming that these people are not interacting with each other in person and they exist in different parts of the world. Yes. Talking exclusively online to each other. Yes. Is that a fair assumption?

Because I I don't agree. Um it's uh it's an assumption that is um I think fair like it can have degrees of fairness uh like it could have degrees of correctness and you can engineer for higher degrees of correctness. Okay. Sure. Sure.

Okay. Um so I think this is um like to me this is like the the point that I want to try to um zoom out here right is like okay so one is uh decentralization is uh like it seems more obviously good at at something that we might call the accountability layer than at something we might call the layer of making individual decisions. two is that uh priv like basic trying to explicitly structure communication is better than uh making like everyone talk to everyone a default. And then three is that making these things work well is like actually a pretty freaking interesting engineering problem. Sorry, which part?

um figuring figuring out these kinds of uh like these kinds of governance structures and figuring out the combinations of them that actually get good results. But if anything, these are to me modules to an existing DAO contract that effectively let people be revoked for uh for their multi-IG capabilities, right? So I mean ultimately like anything is a combination of modules, right? And so, uh, sorry, what I'm trying to say is like to me, if anything, you made the case that it's easier to build it build this as a plugin to an existing DAO contract that is more popular than it is as a full-on brand new replacement. What What is the part of existing Dow contracts that you think is most valuable to keep?

Oh, just the fact that a in a way a vote triggers a change and and ultimately at least 95 out of 100 times Dows are facilitating a flow of funds from one destination to another and everything else is just part of sometimes theater and sometimes uh actual good decision-m and what we just covered is more about okay we we are agreeing to make a decision and it flows the way it does normally but we're adding this future accountability, right? Well, okay. It if we talk about the funds use case, right, then uh like the most vanilla way to do it is you need a vote to approve funds being sent, right? And Right. But it doesn't mean it has to be everybody voting.

You could still be that same person responsible for triggering. Right. Exactly. So what you could do is you could have uh like basically one person responsible for moving funds and then you know you have limits and you have separate mechanisms for adjusting the limits, right? Yeah.

Okay. And so effectively like you still have voting modules but the things that the voting modules are wired to can be very different. And like very plausibly you actually want to take away the wire that connects the uh abil like the voters's ability to just like do like basically have a root key and do whatever they want. Yep. Mhm.

Absolutely. Um so one and I could give one uh example, right? So if we take a D5 protocol then uh one interesting thing about D5 protocols is that often the TVL of the D5 protocol is much greater than the TVL of the um of the D5 protocol token itself usually. Yes. Yes.

And I would argue that this is good and this is necessary because if you do the math in situations where that's not true, the like basically to justify the D5 protocol token being worth more than the TVL like basically the amount of rent it extracts has to be greater than the discount rate and that's like a pretty high amount of extraction. Yeah. So if you but then if you have a protocol where the TVL is much higher than the value of the protocol token then you have a problem which is that someone could buy up 51% of the protocol token and then use that to steal the TVL which has happened on multiple occasions. Yes. And so if you don't want this to happen, then you need like basically mandatory interaction between the uh pro the token holders and uh the thing that's controlling the oracle price.

Yes. Yes. Okay. But but I guess maybe my push back here is um I'm not familiar with a D5 protocol where the DAO is connected to the oracle of the actual D5 contracts. So, right, I mean a lot of them they basically outsource to chain link, right?

Yes. But so what you said is uh I don't think it's a real example is is my what I'm kind of pointing out. What's the example you're most familiar with? No, no. I guess what I'm saying is like almost all DeFi protocols are not tied at all to their Dows, right?

So even if I had a supermajority vote on a decision on what the DAO can dictate, I do not have any path to touch the TVL of the D5 protocol. Uh for a uh is there a Dow that you're familiar with whose oracle is not just pointing to chain link? Technically unis swap's using its own t uh well tap data. Oh yeah, but unis swap isn't um like one of uh it's a good I don't think I know off the top of my head. Right.

Okay. And then you know there's the question of like the internal workings of something like chain link itself, right? Okay. But that's uh but but again that's like you know like one type of use case right and like the use case that's uh like very like operative and top of minds to people is basically the funds allocation use case. Um okay.

So for the uh funds um allocation um use case I guess like what like what are some properties that you think would be valuable for a funds allocation mechanism to have that you think are generally not being met well right now? I don't think that's a big list at all because if anything those things are the the final execution of a decision that's made. So everything usually is how do you get to that decision. M um I do see a lot of actually, you know, that's an interesting one. I do have an answer there, but it kind of comes back to the compliance world over over operational, which is often times if you're giving out a grant to somebody um or for whatever initiative money is being given out um it's almost impossible to have that person receive um the payments in a different chain or in a stable coin because the Dows are incentivized by definition to use their own token.

Um but uh it gets counted against somehow um towards their own success if uh any money going out for any initiative ends up being sold immediately for a stable coin to help increase the value of that protocol. Yeah. No, no, I know the Ethereum Foundation gets this all the time. You know, remember three commandments of the Ethereum Foundation, right? Which you guys are a Dow.

Well, you know, one is um you know, we're D can get a Kraken account. Yeah. Know, it's like one um you know, we're not allowed to sell our ETH and you know, two, we're expected to pay researchers and devs and uh and pay them even more. And uh the Dow has to now take on the financial I guess both sides have to take on the financial risk of of the market price uh affecting you in in mostly a negative case. Yeah.

And so I mean this is um like also like this could possibly be one of those areas again where like privacy actually is the solution cuz like okay imagine if you had a lunar punk dow that just basically automatically you know like gave its grants through a privacy protocol and then the recipient like they would not even know if it was be if they were just like immediately swapping uh swapping it out through like some swap inside of real life. No, it's it's a good way intermediary or you can request a destination to be a a contract that automatically swaps any incoming amount to to a stable coin. Um, but then you get into politics over over tech then people. Yes. Yeah.

Um, so I guess maybe my my kind of question back to you is well do you have a more concrete way to rethink how Dows are managed right now? Are we kind of breaking this by parts and and you're Mhm. coming to the conclusion that all these three cases should be treated differently and we should not really try to blend one solution for all three. Yeah. Yeah.

So I think okay so maybe my conclusions so far would be like one so like one reopen the Overton window right in the sense that like I don't think the starting point should be like start with a coin vote figure out where the coin vote fails and then make a delta. The starting point should be identify the problem and and um engineer around the problem. Um so that's point one. Point two is that privacy matters and we have ZK, we have encryption, we have forms of communication that are not public forums and uh we even have ZK voting frameworks that are getting better and better. Um and so ZK should be much more of a default like we should uh view not just the flow of funds but also flow of information as something that can be engineered around trying to make particular outcomes and uh so that's probably point two right and uh I think um another thing that's uh like really uh also fundamental right is this problem of like allocation of attention like I think we haven't really talked about this Right.

This is the problem where like you know from the perspective of voter actually have you voted in Dallas as much? Uh yeah I've done a couple. Um how I guess uh out of the times that you're theoretically supposed to vote how many times have you voted? Maybe 20% of the time. 20 is high.

Um have you delegated? Uh yes. Yeah. Yeah. Then after that it's after I'm more in favor of a republic model.

Yeah, of course. So, I mean I think like the to me the attention issue is like one of those fundamental things, right? And another way that I think about this is that what we're basically trying to figure out is uh like we're bas basically one part is preference aggregation and um another part is execution, right? And for like often asking people to answer specific decisions is like not actually the best way to like get their information on preferences, right? And uh like separating out preference uh aggregation from uh decision-m more like I think is good, right?

And like and and also you can view accountability as actually being a type of preference aggregation, right? Yeah, it's retroactive, but the accountability is respect is with respect to something, right? And uh like and so it's with respect to some kind of preferences, right? Yeah. Or at least you're maximizing a goal that's agreed upon ahead of time.

Exactly. So one other conclusion I've had is that um I think AI is going to be like actually be pretty important for Dows. Um so actually on the topic of AI, yeah, what kinds of AI crypto projects have you seen so far? Right now I have seen a lot of ways to talk to chat to understand how much crypto I have in my wallet. Um but we're making progress every month.

Uh there have been a lot of creative ways to have um the bots actually aggregate information or sorry LMS aggregate information. And I think kind of the the benefit here um that I feel like is uh is beyond just uh uh requesting information is actually being able to uh create payloads um and parse different information. So I've seen really interesting examples from one of our uh last hackathon which was for just like it was called trifecta that would just like the past ETH global winners were only allowed to compete. Um and uh a cool example there was somebody actually just saying I need X uh token uh and I have all these things in my wallet figure out how do I get the best route and do anywhere from swaps to uh to intent orders and just kept me the outcome and there the bot kind of or just the LM just knew hey these are five strategies I can try and I can aggregate information and make sense of it and then make the best decision. So you can delegate um some of those types of cases.

We have not seen much beyond that. Um there have been a couple projects in the past that like do try to summarize um Dow proposals from forums to kind of more decision uh based uh kind of summaries but um that doesn't help you bet on the quality of a decision. It just helps you maybe save less time um reading it. So I don't have a good uh I don't have a super optimistic opinion on that just yet. Right.

Okay. So one interesting thing there, right, is that like it's actually a good example of this like theme of separating preference expression from execution, right? Because preference expression is basically saying you know like hey I want 100 uh you know like USDC or like I want 0.1 ETH, right? And then execution is well out of all these freaking protocols how do I get it?

Yes. And preferences are a complicated thing, right? So like for example, let's say you have 100 you what you want is 0.1 ETH, but then it turns out that at this particular time the Ethereum chain is going wonky for some reason and literally every means of getting 100 uh or 0.1 ETH would actually involve like losing 40% of slippage.

What what in that world what would you want the AI to do? Abort. Abort. Okay. Now, but in your original request, you did not say abort, did you?

Yes. Because you're because you're assuming that you're assuming the thing you would make the decision like you would as a human and only when you realize you've lost 40% you're like oh I should have actually communicated that right the thing that common law calls the reasonable person test but it's like actually much more subtle than that right because like I think you don't just want it to um optimize for a reasonable person like you really want it to optimize for you. Yeah. Yeah. Yeah.

like okay maybe hypothetical example right let's say you are in um an airport and you need a point to go to uh an ETH global that you literally learned about 1 hour ago and you need to be there tomorrow and this is a high priority need all your credit cards failed and you're buying a plane ticket using crypto and the plane ticket only accepts like say USDC you have ETH and you want the plane ticket now in that particular this if the only way to get it is paying 40% slippage would you wanted to sometimes outcome yeah because you would make a decision based off of getting to that outcome and but let's say you were not mccardic and you were instead someone who like kind of wanted to go to that event but you you don't really care that much they may just be like oh this is then you may just say no right and so this is like the individual case right and so like things that that exist in the individual ual case like they generally have analoges in the organization case, right? And so I think basic like to me like basically this kind of preference um expression is like actually one of the most uh potentially valuable um crypto AI use cases and actually it gets interesting when you start like thinking about preference expression in other domains. Um, so okay, maybe one example. Uh, how how many days have you been in Europe? Five.

How many GDPR clickthroughs have you collect in those five days? Five a day. How how annoyed are you? Uh, 10 out of 10 on that scale. Okay.

Um, so well like the thing that we're trying you're trying to this whole like mess is theoretically trying to do, right, is basically Yeah. You know, like express your preferences in terms of what kind of data you're willing to give up, right? How much meaningful data about your values in terms of your trade-off between data and convenience do you think you've actually communicated to these websites? Near zero. Okay.

So basically now imagine if you know like instead AI was actually involved in that, right? And effectively instead of a very impatient version of yourself doing these negotiations like you actually had an an participating, right? Would you want that? Possibly. Yeah.

Yeah. Yeah. Because if it's doing in the background and I have to be involved. Right. Okay.

So in like one of my guesses also is that in a lot of cases like this is the this is the type of AI D that actually makes sense which is basically a lot of the time you want execution to just be done by AI. Now, why is it an AI Dow? Like what's the Dow part of it? Just the fact that it Well, the Dow part of it in this case would have to be the accountability layer, right? Like say you know you have like $5 million like you don't want uh like literal chat GBT to be able to send $5 million wherever you want.

I understood. Got it. I thought you meant the AI Dow's job is to do these negotiations in this case on your behalf. Right. So the AI D might to do some negotiations, right?

But you know you generally like you want to have limits and you also want to have some accountability layer and for the uh like you want to design some kind of uh accountability layer so that like basically instead of like making a DAO king like you basically you want to be able to remove the D the you know the mad king or at least you want to be able to figure out like which combination of models actually does a good job of satisfying your preferences without like enshrining a single one. Right? Okay. So like this to me is also the well like look this is also for me actually the motivation behind you know the deep funding thing. This is I mean like now there's stuff happening on deep gov right basically you know we want AI but we want AI to be the tool.

We don't want AI to be a king. Yeah. And I guess the the part that I'm happy about in this thing is that we're finally talking about the A in the DAO autonomous. Yeah. I mean remember uh the the original posts from uh back in 2013 that I wrote uh Yes.

Yeah. So, actually maybe as a as a it would be helpful to now kind of have you talk about the motivation of the actual the DAO from 20 uh I guess just walk us through the motivation of the DAO and then Okay. So the Dow Yeah. So the Dow was basically meant to be a like distributed autonomous investment fund, right? a bunch of people pulled their wreath together and then there were like basically people would collectively and like democratically decide on what kinds of things to invest in and like basically this would be a vehicle for like people funding projects together.

So if we were to apply the suggestions we just talked about and retry that experiment, how do you think that'll go? I think it depends entirely on the details. Like maybe we should, right? I think like okay, imagine if we did the DAO 2.0, but you know like number one um like voting is private.

Um number two, you don't just do everything in one layer. You have subdows. And then number three, the subdows themselves are going to be heterogeneous. Some of them will get run by CEOs. Some of them will get run by like pointers to GPT with a few prompts.

Some of them will get run by uh like deepseek running on like some combination of of infra with like a particular fine tune right and some of it might even involve voting in specified groups and then uh like you could imagine uh probably a lot more like layer separation between making decisions and uh if some kind of accountability or preference expression then uh like I could imagine a lot of really cool things happening. Yeah, the reason I brought that up is because that is actually a perfect example where the highest payoff comes usually from a non-conensus decision. Um, and uh, and that kind of counters every other DAO behavior we see now or what we discuss is if you're investing in something, you kind of have to bet on a few years down the road versus six months. So even if somebody thinks, oh, this company did nothing and it's been six months, what are you doing? You made a bad decision, you're fired.

That's probably not how a real life investment would work. Yeah. So, one thing that's interesting is that like I see a bit of a parallel actually to the crypto space. Um, so, okay. So, remember, so the original post I wrote on Dows, they were kind of inspired by Daniel Suarez's book Diamond.

And if you think about what Diamond is like, uh, in in those books, it's actually an owl. It's an autonomous organization that isn't actually that decentralized. Okay. So the analogy is uh if you think about the things that bitcoin is a historical heir to right so you have bitcoin and then what came before bitcoin e-cash e-cash right chiaomi and e-cash right so chom and e-cash it was actually not decentralized at all it was a server but it was private it was private and it protected your privacy even from the server so we basically went from something that was centralized but it satisfied a kind of cipher pod usefulness property to something that was decentralized but that lost that property and then eventually we figured out zk snarks and now we have both. So with Dows, we basically went from, you know, this inspiration, this hypothetical AI thing that was not decentralized and it was autonomous.

Then we went to things that were decentralized that were basically not autonomous at all. And now I'm suggesting, well, what if we actually have both? And I think you want both because you want both. Yeah. You want operational complexity to be reduced, but you also want the ability to uh have more human input in some scenarios that require non-conensus decisions, I guess.

Yeah. Okay. So, so non-con. So, so non-conensus decisions. Uh so, uh would you be more excited about Dows if they were more able to make non-conensus decisions?

Um I don't think it's one thing, right? Like yes, because a lot of to I say that from the context of I think it's fair to say that most Dows here exist to help the underlying protocol be successful long term. Mhm. Um, and that means you have to make decisions that have a much higher payoff for you to be more successful over a period of time. And uh, and sometimes decisions end up being investments that may pay off later in the future.

And you kind of see that it's not a doubt, but like we kind of saw this as a massive initiative with all the the ZK world uh, benefit from the last kind of eight years. So those are things where like, oh, why are you wasting $100 million funding ZK research? uh instead of like making the L1 better or something without realizing that if you make that investment and it does pay off then you end up redeeming yourself or changing trajectory in a much more meaningful way. Um and that's kind of what I mean by non-conensus because some people may say well this seems like a waste of money uh no matter what the decision is and chances are majority of the people will say that because not everybody kind of looks about making decisions for day-to-day or or thinking that far ahead. Um, and you want the ability to not be criticized for a decision that you know will take longer than expected for uh for a future payoff.

Mhm. Okay. Um, if you were um in charge of uh like one of the uh treasuries of a DeFi protocol, then like what would you do? Um, abstractly I would spend as much money as possible trying to integrate with as many things as possible. And um and kind of the reasoning for that is it's in your best interest to treat that as a funnel.

And uh most of DeFi does behave like zero sum in terms of there's a winner take all effect and uh it's in your best interest to get as much reach as possible and uh then double down later on what what's working. Um so be very less um picky about integration initiatives and uh maximize distribution. Would you um like basically spend like most of the treasury very quickly or is sustainability more important? U could be both. I mean it has to be important because you again like this is the long-term bet, right?

If you're saying I'm going to let's say I've seen some Dows invest money in companies as and actual startups. Um there you have to kind of make sure that you're at least thinking three years ahead if not slightly longer. Um, in that world, yes. Um, but you're also f at the same time taking the bet that like, well, maybe if I overspend now and it does help me increase the success of a protocol and potentially the value of my token, um, I may extend the runway. And I would argue that this is exactly what the EF has done.

Um, and and that also is a fair decision. You can actually outspend more than you are able to afford. And uh, and if you think these will work out well, then you end up extending the runway. What about like the metal level like in terms of decision- making authority? Would you wants to keep the decision making authority yourself?

Would you want to like give it to some other structure? Some of both mix of both but but you'll be you would still appoint somebody who is accountable to at least track and observe and and understand how things are going. So so you have in a way a reporting structure to get an answer at any given time what the the progress of your uh your work is. And like what would you want that that reporting structure to be accountable to? What or whom?

What both? Well, whom is in a way I guess if it's a I guess I'm assuming there's probably a normal company behind the DO, right? So that but also fundamentally the community there's there could be an open report. Um but uh the what would be largely just in a way a summary and an activity report of like hey we did this and this is where things are going. So there's also public accountability for who the recipients are.

Mhm. So one type and maybe one example from back in the day, right? So do you remember 10x? Yes. Yeah.

Okay. Remember like back in 2017, I mean I think they were one of the examples, but there were a lot of examples. There were projects that raised a huge amount of money and then they tried a thing and then at some point um you know they were like feeling a bit down and then they effectively pivoted to giving the their founding team like essentially a retirement, right? So, how would you how would you um prevent yourself from doing that? The specific question being how would I prevent a DO from executing a an action that just gives money to or how would Yeah.

How would you prevent a project's leadership from basically just uh like essentially Yeah. Oh. Oh, continuing to feed itself for a very long I see. Yeah. So, so in my example of kind of saying let's try everything and make it a funnel.

Uh you are setting a cap for this is what the initial kind of initiative looks like and I I would only consider deploying more capital to that thing once there's some proof point that it's working because then you have um data points instead of bets without data to uh to see hey this this was a cool thing they got it done in two weeks whereas everybody else took a month and suddenly there's a surge of users from their end and right this is going well let's see if we can continue this. that you have you actually have empirical information. So you' like so like you as an individual would basically empower groups of people but then you'd put like milestones behind. Yes. Um if hypothetically um you know you learned that in two weeks um you know you would have to um permanently retire and let's say you get teleported to a foreign galaxy.

Um what would what structure would you replace yourself with? uh you would try to find the next closest person who believes the same thing that you do and put them in charge. Okay. So you so you'd favor a single person or seems like the premise is that I am that single person in this example. So I would say the best replacement you have, right?

But if you're saying the whole thing is a one person thing. Well, no. I'm saying, you know, you have a D5 project and the yeah, like while you as a founder are um running the thing, then let's say you're very hands-on, but then at some point you have to teleport to a foreign galaxy, then you could replace yourself with a person. You could also replace yourself with a group. I would prefer the person over a group.

Okay. I think I would I think I would go in the opposite direction. I know because there's two co-irectors for Indeed. Indeed there is. Cool.

Um maybe um okay so just uh oh wow it's yeah we should we should we're already 65 wrap this up um but I don't know maybe um so I guess what are some uh so final takeaways from maybe from my side first right is um again that I think you know this uh this whole space of uh you know like this governance mechanisms whether we're talking about um allocating funds um is uh a space or whether we're talking about meeting an Oracle or whether we're talking about like wackier stuff like running a city like re basically burn down the legitimacy of the existing paradigms reopen the Overton window like be you know be more willing to think about things from scratch think about privacy as a first class ingredient I'd also say think about AI as a first class ingredient um I'd also a like multi-level structures are good and so uh like I'm much less of a direct democracy fan than I was 15 years ago. I mean I think thank God. Okay. Um I mean God himself is not a democracy, right? Um thank gods.

Yeah. Um okay. So what what is the word god backwards in Doge? I think exactly. So you added an E at the end, which is good.

Are you familiar with the Yes. Um the selection algorithm for the original Doge? I don't mean Cabo Chen. Uh no. So there's I'm familiar with the Doge of Venice.

No. So Doge is the Italian word for Duke. Just like fun coincidence. Did not realize that. Venice is, you know, a city in Italy.

And uh for about five centuries uh like basically from the 13th century all the way up until Napoleon it had this really interesting mechanism for selecting the duke which was basically you had 1,200 nobles and then those nobles they um elected uh some small number like I think it was like 30 or something like that and then out of that 30 you randomly selected 10 and then those 10 elected like some other number like 45 and then you randomly selected some and then you've elected again you selected some and then finally you elected a group that chose the Doge and this is like very multi level, right? Yeah. This is like a a bisection going on for for an election. Yes. Yeah.

And and what's interesting about this to me, right, is that it's like a very multi-level structure, but it's also a structure that's like very not about entrenching a single party, right? Like actually for someone to be accepted into like in any round, they had to like pass a super majority like twothirds of the vote, right? And so let's like the way that I think about that is like the goal of this like operator is basically you know like you choose like you basically the idea is that people like the people who are competent at the job of governing they themselves are the ones that know the other competent people right and so effectively you're like asking a bunch of people you know who do you know that's the most competent and you ask the same questions to them you ask the same questions to them and then the theory is eventually you know like you find the attractor right And like so this is like actually really freaking cool, right? I agree. This this is a good way to in a way source.

Yeah. No, it's a greatity. So uh basically multi like multilevel is good and uh like b this once you stop thinking about this is multi level on finding the person not on every decision needing this multi-level round. Correct. Yeah.

So this is a one-time action and then this person is in charge until some review community says you are no longer fit. Correct. So yes, this was just for electing the doge. It was not for making individual line items. Um but the um so okay maybe the yeah kind of you know getting back to the takeaway right I think multi level is good and uh like what that also means is that like the design space is much larger than uh you know like I think a lot of people have been thinking and basically you know like go out and be willing to explore.

Um h maybe I mean final question uh for uh for you then is uh what are some things that just like from your point of view if uh either Dows for funding allocation or Dows running things like ENS um were were to do like you would consider to be just like genuinely really interesting and you would feel very excited about I personally would like to see more initiatives around just pure experimentation. Um that kind of just take on a much more long-term view. Um and explicitly kind of do this from the with the premise that we're not trying to optimize the token price right now or it's not a factor. it's about actually improving something longterm which is I don't think I've seen R&D departments for Dows. This is kind of my personal ask for it.

Um and I think the second piece is just uh being able to have them experiment and initiate more um setups where they are allowed to make faster decisions and take more risks with less capital at at kind of a to play with. um just so that they can help quickly refine what good quality looks like. Uh because I I don't think people would disagree with the fact that we've seen a a decent amount of grifting going on with a lot of D proposals. Um and that means you end up discouraging the good players in a lot of cases from trying, but also you don't get to establish a good baseline for what's considered actually exceptional outcome or or delivery of of great work. And um being able to increase the number of outcomes helps you refine and get to the the more refined list of hey look even if this is one person this is what great looks like and you end up uh teaching the whole ecosystem this is how you can set a new norm for what good quality looks like.

Okay. So that sounds like you know more experimentation and particularly find more ways to not be not just mediocre but exceptional. So, sounds like we have a lot of agreement. Yeah, I don't think there's a lot of disagreement. Um, amazing.

So, uh, you know, get out and build. Yep. Well, thank you so much everybody for bearing with us. Uh, please give a massive round of applause for the [Music]