{"id":"4xNlCm5s6ac","title":"Building the Agentic Economy","description":"Jeremy and Marco discuss where we are in the shift towards agentic finance.\n\nAI agents are already trading autonomously, but payments and commerce still need better infrastructure, permissions and guardrails before they can operate safely at scale.\n\nThe intelligence is getting there. Now the infrastructure needs to catch up.\n\nNEAR is the Blockchain for AI\nWebsite: https://near.org/\nNEAR AI: https://near.ai/\nIntents: https://near.org/intents\nChain Abstraction: https://www.proximity.dev/\nSharding: https://nearone.org/\n\nFor the devs\nGitHub: https://github.com/near \nDev Docs: https://docs.near.org/\nDevhub: https://devhub.near.page/ @NEARDevHub\n\nConnect with socials\nTwitter:   / nearprotocol  \nBlog: https://near.org/blog/\nMedium:   / nearprotocol  \n#NEAR #nearprotocol #NEARAI","channel_name":"NEAR Protocol","published_at":"2026-09-28T00:00:00.000Z","thumbnail_url":"https://img.youtube.com/vi/4xNlCm5s6ac/hqdefault.jpg","transcript_text":"We've got Marco and we've got Jeremy. Guys, welcome to the show. You're live. Hi guys. Happy to be here. Great to see you. Great to see you. AI agents are controlling money. I was just telling Ilia and um Alex that I spoke to EMOD from Stability AI, big AI guy who kind of got ousted from SF because I don't know, you know, whatever. And it was, you know, it was kind of when AI was like crypto in 2017 back then. And so what happened is, you know, we got on the show and he's like AI is reaching competent intelligence and like this has happened the last couple of months and it's got to the point where, you know, it's just good enough to be like rival humans and, you know, we're at the era of surpassing human intelligence now. And it's crazy to even say that. And so when we're thinking about agents moving money, you obviously want competent people moving your money or competent agents. So just maybe just like break us down here and I'd love just a quick intro from each of you guys as to where you guys think we truly are at. Set the foundation for AI agents managing money. How realistic? How practical? Just kind of give us the blunt truth and then we can build from there and would love a quick intro. Jeremy, you can start first. For sure. Yeah. Thanks you guys for having us today. Excited to be here. Name's Jeremy, go by it flow on Twitter and then at Near Foundation doing kind of market intelligence. So all things from research side, narrative building, storytelling and then just like overall exploration. Marco, I'll let you introduce yourself then we can hop into the questions. Marco, I've been entrepreneur in education vian web 3 now focus on AI. Um at MetaMask we have just built and released a couple of months ago the MetaMask agent wallet that enables agents to spend and trade autonomously. Beautiful. So maybe just give us a lay of the land gentlemen. Where are we in the era of Agentic Finance? Where are we today? Yeah, I can start us off and hand it off. So, I think still super early days. Uh we haven't really seen a full like autonomous agent loop that's able to spend safely with the proper permissions and like proper signing scheme, stuff like that yet. I think we're close. Uh same thing just like general AI landscape. People are just starting to realize how powerful these tools can be. And if you're already an above average user, you can do really spectacular things. And if you're relatively below average, you'll get up to that average. So these tools as we create them will kind of unlock a lot of just like economic activity hopefully and just make and enrich people's lives. Very cool. Marco, I would clearly divide the uh payments from trading. Uh tradings is definitely more mature. For payments, you need demand and supply, right? demanded ser of services for for agents and supply of services offered by by agents uh and uh we are not there yet in terms of readiness mainly of the demand said that I'm super bullish about what's happening with X42 and MPP uh I'm super interesting um in the micro payments upgrade that X42 did a few month few months ago but if you look at the like overall number as a transaction in the last but compared to December unfortunately we're not there yet and actually some KP is are going down and not up. Totally different stories for for trading. For trading, you don't need like it's like agents directly trading with the products that are already out there. You don't need like a demand and supply. It's working. If you look at the like the number of uh hyperlquid SDK downloads over the last six months is just shocking. So the the direction is clearly more and more programmatic and uh we already have I think the tool to limit and that's what what we will discuss today uh agents in what they're trading and so it the the big challenge is all about like the intelligence of agents in deciding what to trade and the willingness of the human to delegate some authority to those agents. Yeah. And it feels like we're starting to make breakthroughs on both of those fronts. Uh it feels like willingness is becoming a little bit more open. People are seemingly tr and I think they go hand in hand because the models are getting more intelligent. People realize that they are more intelligent and therefore are more willing to give them access and control over some source of funds. And you know, most recently there's been two. Obviously, Astra came out from ChatGBT and OpenAI, but then also Grockbot. I've been seeing so much in in terms of the amount of trust and just pure, you know, admiration for these models. People are trusting them more, realizing how intelligent they are. And from what I've seen, it seems like people are actually starting to make money. They set this thing. uh there's already the infrastructure built for these things to take trades, but now they're actually performing well. H I mean, maybe you guys have seen some internal testing, maybe you guys have tried these things out. When you basically unleash an agent, you give it some money and then you set it free in the wild and you tell it to go take some trades. How much progress has been made? Where were we maybe six months or a year ago in terms of profitability and how much progress has been made up until now? Yeah, I think six months a year ago, we didn't really have like a the formalized word agents per se. Like you had people running bots, which would basically just be like snipers going on pump fun now like pawns and all these places. Those are just kind of programs that were running and they were making money, but you kind of just like set it once and it just does its thing. So that that was a very low-level agent. But now like with Grockbot and like other autonomous kind of trading agents and terminals, you're seeing kind of the proliferation of just set some guard rails and then it just kind of takes the intelligence and goes and just trades. And that's more like the edges you can get is like kind of with your harness and what like directives you give it, what guard rails you give it. So current state now, I think we're definitely early stages, but people are starting to do cool things. People are starting to make a lot of money. So it's just do time and do experimentation to really get it a little bit more mainstream and that will probably come with kind of a formalization of some sort of interface that kind of shows the power of this and like being able to customize it. Yeah. And I know Jeremy you guys are working towards that from the near foundation side right where it's signing permissions. Could you maybe just give us what the POV is from near on what agent permissions should look like, what signing permissions, what guard rail should like if you were just to, you know, give a really, you know, bird's eye view here of like what should be the baseline. Yeah. So there's no like key product that's public available yet, but like in an ideal world, if we could ship it immediately, it would be something where you can kind of move the authority, not just like your assets. And a lot of that goes down to kind of near's core design principles back in the day. Uh like like their key signing signatures and stuff like that. And like most recently we're just passing a new like governance proposal for top level accounts which will right now it's more for like human readable and like humans. But soon it will be kind of like a gentic. So you'll be able to kind of delegate and scope authority out to different agents based off your top level account. You can kind of like en envision different threads of different agents being able to do different things and different functions whether that's for trading or even if you're like another foundation or a business if you want to have like a operations agent that can control all operation spend within like that scoped permission guard rail and then it like doesn't really connect to your top level like [clears throat] authority wise so it can't take your top level money. So it's all about kind of like scoping the permissions to make sure you don't lose your private keys and that all the execution is safe and secure. And then obviously like near is an intentbased system. So that kind of gives you access to all the chains that we have supported which is a little bit broader than kind of all these other like crosschain solutions. Yeah. And then Marco, on the MetaMask side, is is it a similar kind of view here as far as like I know you guys have some really cool X42 and API stuff happening where you guys can kind of let agents operate within some sort of, you know, guideline here. What's your perspective on this permissions, guidelines, guardrails kind of uh perspective here? The kind of problem is not entirely new because delegated authority to other humans has always been around in web 3. So the basic technologies at the end of the day in terms of signing and protection are similar. You don't really want you want to divide the VRP and the key that controls the asset with very scoped keys that can be used by a human or by an agent to do specific things. That's why MetaMask agent wallet has two modalities. One is called guard mode and the other is called beast mode. They are managed by a trusted execution environment. So the agent can't access the assets directly and in in guard mode the agent can only uh operate on like whitelisted protocols and and protected tokens. On this mode uh there is no limitation on the universe of assets that the agents can trade on and the user can also set limits on like specific like limits per day and the request in that case is is escalated to to the human. The needs are very diverse based on user to user I have to say. So there is it's very important to have an extreme granularity. So one of the like most common requests that we're receiving now is to have like custom guards where the the developer wants to trade only on liquid that specific assets. Yeah. All this to to achieve what getting back to your question on like the agent that you you switch it on and it prints money. That's I think like the the wrong framing. The holy grail of like agent created money where while we sleep. I'm sorry. No. Yeah. I'm sorry. Keep it rolling. Keep it rolling. And um for for economical reason of course people have been trying to create algorithms producing money for it. it just it's not the uh while uh if you speak to met MetaMask users the most common uh way to use AI and agents in trading is to get a second opinion and and to form conviction before doing the trade. So the the new flow is that I hear about some alpha from something I read from a friend from like socials from TG and I use my agent to develop conviction of if is it really worth or not. Um and and finally I use the agent and the delegated authority to execute to do some automation and to work less. For example, I decide that I want to play a grid strategy on an asset. I decide the boundaries of the two thresholds and then okay the agent can be autonomous in executing within that strategy. But people are still in control. People still want to think that they are special and that they have a better strategy than somebody else. And uh at the end of the day we feel that we are smartest than our agent in in in taking responsibilities in what we trade and that's fine. Yeah. You know you know what's a like a element of Agentic Finance that has only recently popped up but I I don't think anyone really saw coming was that you know now like Muse just got released by Meta, right? And people are basically telling their Muse agents to go like submit all of these claims for unclaimed property and they're just finding like thousands of dollars that are just out there. And then you know also maybe more notably you know this Navier Stokes problem we were just talking about during the last panel as well. Well that's a millennium math problem. You know they're going to end up receiving a million dollars from that, right? And so obviously they spent a lot more than a million dollars on compute and everything to ultimately you know receive that. But it feels as if there's kind of like these this peripheral edges the e edges of the agentic economy where you can you know don't tell your agent to go trade directly. go tell it to like the most basic example is probably build a business, right? But then it's also like, you know, go try to find unclaimed property. Like go kind of to the edges where there is latent value and try to basically just fill those gaps. And I'm just curious if you guys have come across any other use cases of this that are are kind of like, you know, they're making people money. They're part of the Agentic economy, but they aren't really what people expect when they think of the Agentic financial buildout because it's not directly in finance. It's kind of on the periphery. Is there any any use cases that you guys have found for agents to go out and kind of make money in these more unique ways? Yeah, I I would say it's a really interesting point because it's like everyone has access to this. everyone has access to some level of compute and however much money you have can kind of correlate to the direct like output that you can get. So like when you're scoping for certain things like we can make a cool trading agent that can go trade coins and like make maybe a couple M or you could take a lot of compute go give it to some really cracked researchers and go have them find a new drug or something like that. So I think with kind of this the Savior Stokes problem that kind of is given another green light to the labs that hey we can put a lot of compute behind very specific industries that can create breakthrough technologies that have multi-billion dollar rewards. Yes, it might take a couple hundred million dollars of compute, but it could just be like a shift of focus to those kind of like outside edges that weren't really possible until we had models this intelligent, right? It's kind of like a reconfiguration of the capex. Like, you know, there is speculation that a lot of these labs are going to become like biotech companies. And there was a tweet went pretty viral. It was after this Navier Stokes thing. It's like, oh, you know, I heard there was a rumor Anthropic was pretty close to solving all forms of cancer, right? And they were kind of just they were like making a joke that now OpenAI is going to go, you know, take all of their research and go solve cancer. Like that would be an incredible outcome for these AI models. They will need to start inventing new diseases after that they solve them. Yeah. To to show that they're bad. Yeah. I I think that um it's really a matter of arbitrage. Of course like agents are better than us in collecting signal and come up with opportunities that we couldn't have spotted. But if we all have these tools, this will be true for everybody. Right. Sure. And and same is for how much like intelligence I need to discover that mathematical problem. We are all discussing but there is like a lot of opportunity here for who wants to build this building now wants to be fast. But the opportunity comes from the arbitrage of this not being yet in the hands of everybody or if you're creating a consumer app in making like this agents that aggregate signal uh faster more efficient now. But I long term this like intelligence everywhere will be like so distributed that all this like alpha will be like owned by everybody in a few years. Yeah. So Mark I want to take that back to the top. what you said was this supply and demand problem or just situation if you will where the the commerce side let's say of the agentic finance puzzle is far less developed than the trading side right because I mean we've had quantitative trading down here in in Wall Street for like the last I mean what 30 years right since the since the internet um and so that's happened and now obviously you have like agents trading on chain like this is obvious right cool great but yeah Commerce question based on what you're saying here you know just about kind of this equality of um you know competence or intelligence leading to a kind of level playing field for for everybody who wants to kind of uh participate maybe in markets or in these kind of uh giveaways or or bounties. When you apply that framework to the commerce side like is are you seeing any like spark you know that is going to light the supply and demand equation you know at a steeper slope of adoption or usage like we have level intelligence or sorry we have good intelligence every human has compute they can use it okay so then what so then we're good to go why are we not you know why is this commerce piece still kind of slow The the main blocker is not the intelligence, it's the arness. The connectors between the intelligence and us. Let's make a real example. I want on a Saturday morning to create an agent which is able just to book my restaurant for Saturday night. Incredibly easy, right? And I start with my army with my open claw and I start putting keys to make phone calls. I don't know if you've tried that, but even if you're technical, it takes like six hours and it doesn't work. It doesn't work smoothly, right? So, but it's the problem is not the model. the model I can solve like as we just discussed mathematica. So and this is changing every month and it's improving a lot. So how much it's easy for a consumer per for a consumer that has no technical background to create an agent that is able to book a hotel to to buy some sneakers. Let's start from the basics what people do. This is still not easy enough. So that's the main blocker to me. The sec second um it's still the cost. So if you do the math on like the impact on a consumer how much it costs to to do all this with tokens all day long it's they would like spend thousands of dollars of tokens that is not like so the velocity of token per second the kura velocity of token per second and the the cur of cost per token that is like very steep will improve a lot. I'm pretty bullish that in a few months uh both problems will be largely mitigated. Yeah. Jeremy, anything to add there? Yeah, I think more maybe on the the trading side. I think we're seeing a rise in social trading with with FOMO and stuff like that. So, I think it's cool and it's a really cool paradigm. But I think we'll also see an increase in like spoofing as well because now you can put out kind of fake data. like we already see people aird dropping small even sometimes large amounts of tokens to some of the top traders and then that kind of like pumps up the chart and goes back down. So we also need to come up with like certain types of guard rails to protect consumers and decision makers and agents as well because when you think about even like when cloud code came out there used to be like you have to like you have to accept each change you want to make right and then they put oh now you can just auto accept all changes and basically remove every guardrail possible and now you just have some machine going crazy on your machine that's not the safest so we need thoughtful design again uh I know it's super easy to make software these days, but it's almost more important to take a lot of time and consider what you're designing, how people are going to use it, and like what are the edge cases of how people can then go and destroy it because now they can just prompt an agent to go try to destroy your software. And if you didn't give that thought, you're just kind of screwed. Yeah. Yeah. Yeah. I mean there the implications of this feel like they're going to be farreaching both on the you know security side of things as well as you know the the social trading composability side of things. Um Marco just curious you know how you guys are approaching this from the wallet perspective given that you know this is kind of a new paradigm you guys have released new products before. How do you balance innovation and security? Yeah. And also just given the the the consensus news as well this week, it's uh I don't know if it affects anything the user doing. One thing is scoping the what the agent can do and another problem is defending the agent from doing things that are within the scope but shouldn't happen because like the counterpart convince the agent to do that. So these are two different problems. the the wallet tackles the the former. How can you limit the scope of an agent? Security for metam mask is like number one priority always. It comes even before innovation. Uh it's at the end of the day with self-cast security of the asset is the most important thing and the the the agent doesn't have access to the assets themselves that are in a TE. All the access is always scoped with like two different levels as we were discussing. you can limit the access and I think that 7702 that has been released uh 6 months ago has really helped in general the entire industry because you can just pick a new way keep the key of AOA out of a like not in the agent so please don't give your agent your private key and then you can temporarily like evolve the EOA in a smart contract with all the custom dynamic that you want also using a smart contract and and that's it if you don't like TE's because as we know like this is a general term that we use often but the security of the is really depends on which e architecture we we we are talking about. So this is the wallet part but then there is another part that other players are tackling which is how can I create an output guard that if the agent sends a request that is allowed from the wallet but is against what's happening into the conversation. Let's say that in the system prompt that is written don't refund clients from Europe for some compliance reason the wallet doesn't read the conversation right it just received the request so you need output gar that are deterministically on the hotspot in less than 100 millisecond able to read the conversation and stop the agent and that's like another important problem uh uh to to answer your question on on the consistent news of this week so yeah uh the protocol those group of consensus that run all our like uh Ethereum efforts are not part of consensus anymore and has been spun out in a new company that is called consensus while the current consensus that so MetaMask basically is renamed to to MetaMask. So it's a straightforward way to align to users with what like the brand that people uh have known and loved. Yeah. Yeah. Awesome. Well guys, this is uh as fascinating is as it is eerie. I mean, how about that guy from Anthropic the other day, crashing out on the timeline two months before his vesting cliff start or ended or whatever. So, not even trying to like, you know, vesting cliff ended, got rich, get out of here. Just before it even started or, you know, ended, he's just like, I'm out and this is going to kill people. It's just like, whoa. Okay. I mean, the hell's going on? What the hell's going on in Silicon Valley, guys? too much acid. [snorts] It is scary, but it's good. I mean, every innovation comes with ups and downs, right? And people are afraid to change. We have to make sure it's thoughtfully designed and people are thinking about it properly. We're all humans. We'll make errors, but hopefully we make more good decisions than bad ones. I love it. Well, guys, looking forward to the future of the agentic finance world that we're entering that you guys are building out. I know we're a little past time. Thank you for your time, Jeremy and Marco. awesome session and look forward to seeing what you guys can cook. Have a good one. Bye everybody. You too.","transcript_synced_at":"2026-11-03T04:19:07.314Z","segments":[{"start_seconds":0,"text":"We've got Marco and we've got Jeremy. Guys, welcome to the show. You're live. Hi guys. Happy to be here.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=0s"},{"start_seconds":6,"text":"Great to see you. Great to see you. AI agents are controlling money. I was just telling Ilia and um Alex that I spoke to","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=6s"},{"start_seconds":14,"text":"EMOD from Stability AI, big AI guy who kind of got ousted from SF because I don't know, you know, whatever. And it","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=14s"},{"start_seconds":22,"text":"was, you know, it was kind of when AI was like crypto in 2017 back then. And so what happened is, you know, we got on the show and he's like AI is reaching","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=22s"},{"start_seconds":31,"text":"competent intelligence and like this has happened the last couple of months and it's got to the point where, you know, it's just good enough to be like rival","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=31s"},{"start_seconds":39,"text":"humans and, you know, we're at the era of surpassing human intelligence now.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=39s"},{"start_seconds":43,"text":"And it's crazy to even say that. And so when we're thinking about agents moving money, you obviously want competent","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=43s"},{"start_seconds":51,"text":"people moving your money or competent agents. So just maybe just like break us down here and I'd love just a quick intro from each of you guys as to where","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=51s"},{"start_seconds":59,"text":"you guys think we truly are at. Set the foundation for AI agents managing money.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=59s"},{"start_seconds":64,"text":"How realistic? How practical? Just kind of give us the blunt truth and then we can build from there and would love a quick intro. Jeremy, you can start first.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=64s"},{"start_seconds":73,"text":"For sure. Yeah. Thanks you guys for having us today. Excited to be here.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=73s"},{"start_seconds":76,"text":"Name's Jeremy, go by it flow on Twitter and then at Near Foundation doing kind of market intelligence. So all things from research side, narrative building,","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=76s"},{"start_seconds":85,"text":"storytelling and then just like overall exploration. Marco, I'll let you introduce yourself then we can hop into the questions.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=85s"},{"start_seconds":93,"text":"Marco, I've been entrepreneur in education vian web 3 now focus on AI. Um at MetaMask we have just built and released a couple of months ago the","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=93s"},{"start_seconds":101,"text":"MetaMask agent wallet that enables agents to spend and trade autonomously.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=101s"},{"start_seconds":108,"text":"Beautiful. So maybe just give us a lay of the land gentlemen. Where are we in the era of Agentic Finance? Where are we today?","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=108s"},{"start_seconds":115,"text":"Yeah, I can start us off and hand it off. So, I think still super early days.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=115s"},{"start_seconds":119,"text":"Uh we haven't really seen a full like autonomous agent loop that's able to spend safely with the proper permissions and like proper signing scheme, stuff","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=119s"},{"start_seconds":126,"text":"like that yet. I think we're close. Uh same thing just like general AI landscape. People are just starting to realize how powerful these tools can be.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=126s"},{"start_seconds":134,"text":"And if you're already an above average user, you can do really spectacular things. And if you're relatively below average, you'll get up to that average.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=134s"},{"start_seconds":142,"text":"So these tools as we create them will kind of unlock a lot of just like economic activity hopefully and just make and enrich people's lives.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=142s"},{"start_seconds":153,"text":"Very cool. Marco, I would clearly divide the uh payments from trading. Uh tradings is definitely more mature. For payments, you need","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=153s"},{"start_seconds":161,"text":"demand and supply, right? demanded ser of services for for agents and supply of services offered by by agents uh and uh","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=161s"},{"start_seconds":169,"text":"we are not there yet in terms of readiness mainly of the demand said that I'm super bullish about what's happening with X42 and MPP uh I'm super","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=169s"},{"start_seconds":178,"text":"interesting um in the micro payments upgrade that X42 did a few month few months ago but if you look at the like","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=178s"},{"start_seconds":186,"text":"overall number as a transaction in the last but compared to December unfortunately we're not there yet and actually some KP is are going down and","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=186s"},{"start_seconds":193,"text":"not up. Totally different stories for for trading. For trading, you don't need like it's like agents directly trading with the products that are already out","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=193s"},{"start_seconds":201,"text":"there. You don't need like a demand and supply. It's working. If you look at the like the number of uh hyperlquid SDK","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=201s"},{"start_seconds":209,"text":"downloads over the last six months is just shocking. So the the direction is clearly more and more programmatic and","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=209s"},{"start_seconds":216,"text":"uh we already have I think the tool to limit and that's what what we will discuss today uh agents in what they're","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=216s"},{"start_seconds":222,"text":"trading and so it the the big challenge is all about like the intelligence of agents in deciding what to trade and the","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=222s"},{"start_seconds":231,"text":"willingness of the human to delegate some authority to those agents.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=231s"},{"start_seconds":235,"text":"Yeah. And it feels like we're starting to make breakthroughs on both of those fronts. Uh it feels like willingness is","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=235s"},{"start_seconds":243,"text":"becoming a little bit more open. People are seemingly tr and I think they go hand in hand because the models are getting more intelligent. People realize","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=243s"},{"start_seconds":252,"text":"that they are more intelligent and therefore are more willing to give them access and control over some source of funds. And you know, most recently","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=252s"},{"start_seconds":260,"text":"there's been two. Obviously, Astra came out from ChatGBT and OpenAI, but then also Grockbot. I've been seeing so much","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=260s"},{"start_seconds":267,"text":"in in terms of the amount of trust and just pure, you know, admiration for these models. People are trusting them","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=267s"},{"start_seconds":275,"text":"more, realizing how intelligent they are. And from what I've seen, it seems like people are actually starting to make money. They set this thing. uh","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=275s"},{"start_seconds":282,"text":"there's already the infrastructure built for these things to take trades, but now they're actually performing well.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=282s"},{"start_seconds":290,"text":"H I mean, maybe you guys have seen some internal testing, maybe you guys have tried these things out. When you basically unleash an agent, you give it","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=290s"},{"start_seconds":298,"text":"some money and then you set it free in the wild and you tell it to go take some trades. How much progress has been made?","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=298s"},{"start_seconds":304,"text":"Where were we maybe six months or a year ago in terms of profitability and how much progress has been made up until now?","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=304s"},{"start_seconds":312,"text":"Yeah, I think six months a year ago, we didn't really have like a the formalized word agents per se. Like you had people running bots, which would basically just","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=312s"},{"start_seconds":321,"text":"be like snipers going on pump fun now like pawns and all these places. Those are just kind of programs that were running and they were making money, but","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=321s"},{"start_seconds":329,"text":"you kind of just like set it once and it just does its thing. So that that was a very low-level agent. But now like with Grockbot and like other autonomous kind","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=329s"},{"start_seconds":337,"text":"of trading agents and terminals, you're seeing kind of the proliferation of just set some guard rails and then it just kind of takes the intelligence and goes","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=337s"},{"start_seconds":346,"text":"and just trades. And that's more like the edges you can get is like kind of with your harness and what like directives you give it, what guard rails","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=346s"},{"start_seconds":353,"text":"you give it. So current state now, I think we're definitely early stages, but people are starting to do cool things.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=353s"},{"start_seconds":359,"text":"People are starting to make a lot of money. So it's just do time and do experimentation to really get it a little bit more mainstream and that will probably come with kind of a","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=359s"},{"start_seconds":367,"text":"formalization of some sort of interface that kind of shows the power of this and like being able to customize it.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=367s"},{"start_seconds":374,"text":"Yeah. And I know Jeremy you guys are working towards that from the near foundation side right where it's signing permissions. Could you maybe just give","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=374s"},{"start_seconds":382,"text":"us what the POV is from near on what agent permissions should look like, what signing permissions, what guard rail","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=382s"},{"start_seconds":390,"text":"should like if you were just to, you know, give a really, you know, bird's eye view here of like what should be the baseline.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=390s"},{"start_seconds":397,"text":"Yeah. So there's no like key product that's public available yet, but like in an ideal world, if we could ship it immediately, it would be something where","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=397s"},{"start_seconds":406,"text":"you can kind of move the authority, not just like your assets. And a lot of that goes down to kind of near's core design principles back in the day. Uh like like","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=406s"},{"start_seconds":414,"text":"their key signing signatures and stuff like that. And like most recently we're just passing a new like governance proposal for top level accounts which","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=414s"},{"start_seconds":422,"text":"will right now it's more for like human readable and like humans. But soon it will be kind of like a gentic. So you'll be able to kind of delegate and scope","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=422s"},{"start_seconds":430,"text":"authority out to different agents based off your top level account. You can kind of like en envision different threads of different agents being able to do different things and different functions","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=430s"},{"start_seconds":439,"text":"whether that's for trading or even if you're like another foundation or a business if you want to have like a operations agent that can control all","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=439s"},{"start_seconds":446,"text":"operation spend within like that scoped permission guard rail and then it like doesn't really connect to your top level like [clears throat] authority wise so","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=446s"},{"start_seconds":454,"text":"it can't take your top level money. So it's all about kind of like scoping the permissions to make sure you don't lose your private keys and that all the","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=454s"},{"start_seconds":462,"text":"execution is safe and secure. And then obviously like near is an intentbased system. So that kind of gives you access to all the chains that we have supported","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=462s"},{"start_seconds":470,"text":"which is a little bit broader than kind of all these other like crosschain solutions.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=470s"},{"start_seconds":474,"text":"Yeah. And then Marco, on the MetaMask side, is is it a similar kind of view here as far as like I know you guys have some really cool X42 and API stuff","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=474s"},{"start_seconds":482,"text":"happening where you guys can kind of let agents operate within some sort of, you know, guideline here. What's your perspective on this permissions,","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=482s"},{"start_seconds":490,"text":"guidelines, guardrails kind of uh perspective here?","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=490s"},{"start_seconds":495,"text":"The kind of problem is not entirely new because delegated authority to other humans has always been around in web 3.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=495s"},{"start_seconds":501,"text":"So the basic technologies at the end of the day in terms of signing and protection are similar. You don't really want you want to divide the VRP and the","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=501s"},{"start_seconds":509,"text":"key that controls the asset with very scoped keys that can be used by a human or by an agent to do specific things.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=509s"},{"start_seconds":515,"text":"That's why MetaMask agent wallet has two modalities. One is called guard mode and the other is called beast mode. They are managed by a trusted execution","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=515s"},{"start_seconds":524,"text":"environment. So the agent can't access the assets directly and in in guard mode","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=524s"},{"start_seconds":531,"text":"the agent can only uh operate on like whitelisted protocols and and protected tokens. On this mode uh there is no","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=531s"},{"start_seconds":540,"text":"limitation on the universe of assets that the agents can trade on and the user can also set limits on like","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=540s"},{"start_seconds":547,"text":"specific like limits per day and the request in that case is is escalated to to the human. The needs are very diverse","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=547s"},{"start_seconds":555,"text":"based on user to user I have to say. So there is it's very important to have an extreme granularity. So one of the like most common requests that we're","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=555s"},{"start_seconds":562,"text":"receiving now is to have like custom guards where the the developer wants to trade only on liquid that specific assets.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=562s"},{"start_seconds":570,"text":"Yeah.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=570s"},{"start_seconds":570,"text":"All this to to achieve what getting back to your question on like the agent that you you switch it on and it prints money. That's I think like the the wrong","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=570s"},{"start_seconds":578,"text":"framing. The holy grail of like agent created money where while we sleep. I'm sorry. No.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=578s"},{"start_seconds":589,"text":"Yeah. I'm sorry. Keep it rolling. Keep it rolling.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=589s"},{"start_seconds":590,"text":"And um for for economical reason of course people have been trying to create algorithms producing money for it. it just it's not the uh while uh if you","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=590s"},{"start_seconds":599,"text":"speak to met MetaMask users the most common uh way to use AI and agents in","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=599s"},{"start_seconds":605,"text":"trading is to get a second opinion and and to form conviction before doing the trade. So the the new flow is that I","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=605s"},{"start_seconds":614,"text":"hear about some alpha from something I read from a friend from like socials from TG and I use my agent to develop","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=614s"},{"start_seconds":621,"text":"conviction of if is it really worth or not. Um and and finally I use the agent and the delegated authority to execute to do some automation and to work less.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=621s"},{"start_seconds":631,"text":"For example, I decide that I want to play a grid strategy on an asset. I decide the boundaries of the two thresholds and then okay the agent can","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=631s"},{"start_seconds":640,"text":"be autonomous in executing within that strategy. But people are still in control. People still want to think that they are special and that they have a","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=640s"},{"start_seconds":648,"text":"better strategy than somebody else. And uh at the end of the day we feel that we are smartest than our agent in in in taking responsibilities in what we trade and that's fine.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=648s"},{"start_seconds":659,"text":"Yeah.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=659s"},{"start_seconds":659,"text":"You know you know what's a like a element of Agentic Finance that has only recently popped up but I I don't think anyone really saw coming was that you","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=659s"},{"start_seconds":667,"text":"know now like Muse just got released by Meta, right? And people are basically telling their Muse agents to go like","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=667s"},{"start_seconds":677,"text":"submit all of these claims for unclaimed property and they're just finding like thousands of dollars that are just out there. And then you know also maybe more","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=677s"},{"start_seconds":686,"text":"notably you know this Navier Stokes problem we were just talking about during the last panel as well. Well that's a millennium math problem. You","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=686s"},{"start_seconds":693,"text":"know they're going to end up receiving a million dollars from that, right? And so obviously they spent a lot more than a","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=693s"},{"start_seconds":701,"text":"million dollars on compute and everything to ultimately you know receive that. But it feels as if there's kind of like these this peripheral edges","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=701s"},{"start_seconds":710,"text":"the e edges of the agentic economy where you can you know don't tell your agent to go trade directly. go tell it to like","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=710s"},{"start_seconds":720,"text":"the most basic example is probably build a business, right? But then it's also like, you know, go try to find unclaimed property. Like go kind of to the edges","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=720s"},{"start_seconds":728,"text":"where there is latent value and try to basically just fill those gaps. And I'm just curious if you guys have come","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=728s"},{"start_seconds":735,"text":"across any other use cases of this that are are kind of like, you know, they're making people money. They're part of the Agentic economy, but they aren't really","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=735s"},{"start_seconds":743,"text":"what people expect when they think of the Agentic financial buildout because it's not directly in finance. It's kind of on the periphery. Is there any any","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=743s"},{"start_seconds":751,"text":"use cases that you guys have found for agents to go out and kind of make money in these more unique ways?","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=751s"},{"start_seconds":758,"text":"Yeah, I I would say it's a really interesting point because it's like everyone has access to this. everyone has access to some level of compute and","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=758s"},{"start_seconds":766,"text":"however much money you have can kind of correlate to the direct like output that you can get. So like when you're scoping for certain things like we can make a","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=766s"},{"start_seconds":774,"text":"cool trading agent that can go trade coins and like make maybe a couple M or you could take a lot of compute go give it to some really cracked","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=774s"},{"start_seconds":782,"text":"researchers and go have them find a new drug or something like that. So I think with kind of this the Savior","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=782s"},{"start_seconds":790,"text":"Stokes problem that kind of is given another green light to the labs that hey we can put a lot of compute behind very specific industries that can create","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=790s"},{"start_seconds":797,"text":"breakthrough technologies that have multi-billion dollar rewards. Yes, it might take a couple hundred million","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=797s"},{"start_seconds":804,"text":"dollars of compute, but it could just be like a shift of focus to those kind of like outside edges that weren't really","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=804s"},{"start_seconds":812,"text":"possible until we had models this intelligent, right? It's kind of like a reconfiguration of the capex. Like, you","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=812s"},{"start_seconds":820,"text":"know, there is speculation that a lot of these labs are going to become like biotech companies. And there was a tweet went pretty viral. It was after this","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=820s"},{"start_seconds":827,"text":"Navier Stokes thing. It's like, oh, you know, I heard there was a rumor Anthropic was pretty close to solving all forms of cancer, right? And they","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=827s"},{"start_seconds":835,"text":"were kind of just they were like making a joke that now OpenAI is going to go, you know, take all of their research and go solve cancer. Like that would be an incredible outcome for these AI models.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=835s"},{"start_seconds":847,"text":"They will need to start inventing new diseases after that they solve them.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=847s"},{"start_seconds":850,"text":"Yeah. To to show that they're bad. Yeah. I I think that um it's really a matter of arbitrage. Of course like agents are better than us in collecting","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=850s"},{"start_seconds":859,"text":"signal and come up with opportunities that we couldn't have spotted. But if we all have these tools, this will be true for everybody. Right.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=859s"},{"start_seconds":866,"text":"Sure. And and same is for how much like intelligence I need to discover that mathematical problem. We are all discussing but there is like a lot of","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=866s"},{"start_seconds":873,"text":"opportunity here for who wants to build this building now wants to be fast. But the opportunity comes from the arbitrage of this not being yet in the hands of","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=873s"},{"start_seconds":882,"text":"everybody or if you're creating a consumer app in making like this agents that aggregate signal uh faster more","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=882s"},{"start_seconds":889,"text":"efficient now. But I long term this like intelligence everywhere will be like so distributed that all this like alpha","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=889s"},{"start_seconds":897,"text":"will be like owned by everybody in a few years.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=897s"},{"start_seconds":900,"text":"Yeah. So Mark I want to take that back to the top. what you said was this supply and demand problem or just situation if you will where the the","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=900s"},{"start_seconds":908,"text":"commerce side let's say of the agentic finance puzzle is far less developed than the trading side right because I","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=908s"},{"start_seconds":916,"text":"mean we've had quantitative trading down here in in Wall Street for like the last I mean what 30 years right since the","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=916s"},{"start_seconds":923,"text":"since the internet um and so that's happened and now obviously you have like agents trading on chain like this is obvious right cool great but yeah","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=923s"},{"start_seconds":932,"text":"Commerce question based on what you're saying here you know just about kind of this equality of um you know competence","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=932s"},{"start_seconds":940,"text":"or intelligence leading to a kind of level playing field for for everybody who wants to kind of uh participate","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=940s"},{"start_seconds":947,"text":"maybe in markets or in these kind of uh giveaways or or bounties. When you apply that framework to the commerce side like","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=947s"},{"start_seconds":955,"text":"is are you seeing any like spark you know that is going to light the supply and demand equation you know at a","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=955s"},{"start_seconds":963,"text":"steeper slope of adoption or usage like we have level intelligence or sorry we have good intelligence every human has","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=963s"},{"start_seconds":971,"text":"compute they can use it okay so then what so then we're good to go why are we not you know why is this commerce piece","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=971s"},{"start_seconds":978,"text":"still kind of slow The the main blocker is not the intelligence, it's the arness. The connectors between the intelligence and","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=978s"},{"start_seconds":986,"text":"us. Let's make a real example. I want on a Saturday morning to create an agent which is able just to book my restaurant for Saturday night. Incredibly easy,","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=986s"},{"start_seconds":993,"text":"right? And I start with my army with my open claw and I start putting keys to make phone calls. I don't know if you've tried that, but even if you're","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=993s"},{"start_seconds":1001,"text":"technical, it takes like six hours and it doesn't work. It doesn't work smoothly, right? So, but it's the problem is not the model. the model I","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1001s"},{"start_seconds":1009,"text":"can solve like as we just discussed mathematica. So and this is changing every month and it's improving a lot. So","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1009s"},{"start_seconds":1017,"text":"how much it's easy for a consumer per for a consumer that has no technical background to create an agent that is","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1017s"},{"start_seconds":1025,"text":"able to book a hotel to to buy some sneakers. Let's start from the basics what people do. This is still not easy","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1025s"},{"start_seconds":1032,"text":"enough. So that's the main blocker to me. The sec second um it's still the cost. So if you do the math on like the","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1032s"},{"start_seconds":1040,"text":"impact on a consumer how much it costs to to do all this with tokens all day long it's they would like spend","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1040s"},{"start_seconds":1048,"text":"thousands of dollars of tokens that is not like so the velocity of token per second the kura velocity of token per","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1048s"},{"start_seconds":1054,"text":"second and the the cur of cost per token that is like very steep will improve a lot. I'm pretty bullish that in a few","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1054s"},{"start_seconds":1061,"text":"months uh both problems will be largely mitigated. Yeah. Jeremy, anything to add there?","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1061s"},{"start_seconds":1069,"text":"Yeah, I think more maybe on the the trading side. I think we're seeing a rise in social trading with with FOMO and stuff like that. So, I think it's cool and it's a really cool paradigm.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1069s"},{"start_seconds":1080,"text":"But I think we'll also see an increase in like spoofing as well because now you can put out kind of fake data. like we already see people aird dropping small","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1080s"},{"start_seconds":1088,"text":"even sometimes large amounts of tokens to some of the top traders and then that kind of like pumps up the chart and goes back down. So we also need to come up","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1088s"},{"start_seconds":1096,"text":"with like certain types of guard rails to protect consumers and decision makers and agents as well because when you","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1096s"},{"start_seconds":1103,"text":"think about even like when cloud code came out there used to be like you have to like you have to accept each change you want to make right and then they put","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1103s"},{"start_seconds":1111,"text":"oh now you can just auto accept all changes and basically remove every guardrail possible and now you just have some machine going crazy on your machine","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1111s"},{"start_seconds":1120,"text":"that's not the safest so we need thoughtful design again uh I know it's super easy to make software these days, but it's almost more important to take a","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1120s"},{"start_seconds":1128,"text":"lot of time and consider what you're designing, how people are going to use it, and like what are the edge cases of how people can then go and destroy it because now they can just prompt an","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1128s"},{"start_seconds":1136,"text":"agent to go try to destroy your software. And if you didn't give that thought, you're just kind of screwed.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1136s"},{"start_seconds":1142,"text":"Yeah. Yeah. Yeah. I mean there the implications of this feel like they're going to be farreaching both on the you","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1142s"},{"start_seconds":1150,"text":"know security side of things as well as you know the the social trading composability side of things. Um Marco just curious you know how you guys are","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1150s"},{"start_seconds":1158,"text":"approaching this from the wallet perspective given that you know this is kind of a new paradigm you guys have released new products before. How do you balance innovation and security? Yeah.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1158s"},{"start_seconds":1169,"text":"And also just given the the the consensus news as well this week, it's uh I don't know if it affects anything the user doing.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1169s"},{"start_seconds":1180,"text":"One thing is scoping the what the agent can do and another problem is defending the agent from doing things that are","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1180s"},{"start_seconds":1188,"text":"within the scope but shouldn't happen because like the counterpart convince the agent to do that. So these are two different problems. the the wallet","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1188s"},{"start_seconds":1197,"text":"tackles the the former. How can you limit the scope of an agent? Security for metam mask is like number one","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1197s"},{"start_seconds":1204,"text":"priority always. It comes even before innovation. Uh it's at the end of the day with self-cast security of the asset","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1204s"},{"start_seconds":1211,"text":"is the most important thing and the the the agent doesn't have access to the assets themselves that are in a TE. All","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1211s"},{"start_seconds":1219,"text":"the access is always scoped with like two different levels as we were discussing. you can limit the access and","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1219s"},{"start_seconds":1226,"text":"I think that 7702 that has been released uh 6 months ago has really helped in general the entire industry because you can just pick a new way keep the key of","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1226s"},{"start_seconds":1235,"text":"AOA out of a like not in the agent so please don't give your agent your private key and then you","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1235s"},{"start_seconds":1242,"text":"can temporarily like evolve the EOA in a smart contract with all the custom dynamic that you want also using a smart contract and and that's it if you don't","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1242s"},{"start_seconds":1251,"text":"like TE's because as we know like this is a general term that we use often but the security of the is really depends on which e architecture we we we are","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1251s"},{"start_seconds":1261,"text":"talking about. So this is the wallet part but then there is another part that other players are tackling which is how","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1261s"},{"start_seconds":1267,"text":"can I create an output guard that if the agent sends a request that is allowed from the wallet but is against what's happening","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1267s"},{"start_seconds":1276,"text":"into the conversation. Let's say that in the system prompt that is written don't refund clients from Europe for some compliance reason the wallet doesn't","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1276s"},{"start_seconds":1286,"text":"read the conversation right it just received the request so you need output gar that are deterministically on the hotspot in less than 100 millisecond","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1286s"},{"start_seconds":1294,"text":"able to read the conversation and stop the agent and that's like another important problem uh uh to to answer your question on on the consistent news","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1294s"},{"start_seconds":1303,"text":"of this week so yeah uh the protocol those group of consensus that run all our like uh Ethereum","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1303s"},{"start_seconds":1311,"text":"efforts are not part of consensus anymore and has been spun out in a new company that is called consensus while the current consensus that so","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1311s"},{"start_seconds":1319,"text":"MetaMask basically is renamed to to MetaMask. So it's a straightforward way to align","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1319s"},{"start_seconds":1327,"text":"to users with what like the brand that people uh have known and loved.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1327s"},{"start_seconds":1334,"text":"Yeah. Yeah. Awesome. Well guys, this is uh as fascinating is as it is eerie. I mean, how about that guy from Anthropic","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1334s"},{"start_seconds":1342,"text":"the other day, crashing out on the timeline two months before his vesting cliff start or ended or whatever. So, not even trying to like, you know,","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1342s"},{"start_seconds":1350,"text":"vesting cliff ended, got rich, get out of here. Just before it even started or, you know, ended, he's just like, I'm out","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1350s"},{"start_seconds":1357,"text":"and this is going to kill people. It's just like, whoa. Okay. I mean, the hell's going on? What the hell's going on in Silicon Valley, guys? too much acid.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1357s"},{"start_seconds":1367,"text":"[snorts]","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1367s"},{"start_seconds":1367,"text":"It is scary, but it's good. I mean, every innovation comes with ups and downs, right? And people are afraid to change. We have to make sure it's","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1367s"},{"start_seconds":1374,"text":"thoughtfully designed and people are thinking about it properly.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1374s"},{"start_seconds":1378,"text":"We're all humans. We'll make errors, but hopefully we make more good decisions than bad ones.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1378s"},{"start_seconds":1383,"text":"I love it. Well, guys, looking forward to the future of the agentic finance world that we're entering that you guys are building out. I know we're a little past time. Thank you for your time,","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1383s"},{"start_seconds":1392,"text":"Jeremy and Marco. awesome session and look forward to seeing what you guys can cook. Have a good one. Bye everybody.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1392s"},{"start_seconds":1399,"text":"You too.","url":"https://www.youtube.com/watch?v=4xNlCm5s6ac&t=1399s"}],"youtube_url":"https://youtube.com/watch?v=4xNlCm5s6ac","url":"https://near-pedia.xyz/videos/4xNlCm5s6ac"}