New Ethereum talks, every Monday. The week's conference uploads by event, in your inbox.

Loading player…

Agents With Allowances, Embedded Wallet Tools For A New Economy I Itai Turbahn at Pragma Denver 2025

ETHGlobalThu, Oct 9, 2025, 12:00 AM

Join Itai Turbahn, Co-Founder of Dynamic, at Pragma Denver 2025 as he introduces a new approach to AI Agents with Allowances, powered by TSS (Threshold Signature) MPC and next-gen embedded wallet tools. Discover how Dynamic’s flexible key management solution securely grants AI agents fine-grained spending permissions, automates on-chain tasks, and supports both “human-in-the-loop” and fully autonomous models. Learn why rule engines, ephemeral key shares, and EIP-7702 gasless transactions play a critical role in creating a trustless environment for everyday use cases—ranging from booking travel to trading. If you’re exploring the future of AI, Web3 automation, and advanced wallet orchestration, this talk uncovers the building blocks for a new, decentralized economy. 00:00 Intro & Agents with Allowances 01:17 Dynamic’s Embedded Wallets & Key Management 02:23 Focusing on Human Onboarding & Industry Partners 03:27 AI Agents: The Next Frontier of Automation 05:08 TSS MPC Announcement & Secure Key Sharing 08:04 Rule Engine & Fine-Grained Spending Policies 09:10 Semi-Autonomous vs. Fully Autonomous Agents 14:01 No Human in the Loop: Server-Only Wallet Flow 15:03 EIP-7702 & Gasless Transactions 16:56 Conclusion & Future of AI Wallet Tools _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 🇺🇸 *Pragma Denver* Pragma Denver 2025 was held on February 25th at ReelWorks in Denver, Colorado and was an in-person summit for builders and leaders in the web3 ecosystem. Watch the full Pragma Denver YouTube Playlist here: https://youtube.com/playlist?list=PLXzKMXK2aHh6mvMfVn2gIK2uhRiTHjIg9&si=D7XfnBKvZL7sthbU ETHGlobal's Pragma series takes place in cities around the world, and is designed to be a different kind of event. Pragma is a one-stage conference with founders-only on stage, bringing together a small group of curated attendees and speakers to discuss the future of web3 and reflect on the past. _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ ✅ *Follow Itai Turbahn* X: https://x.com/turbahn ✅ Follow *Dynamic* X: https://x.com/dynamic_xyz ✅ *Follow ETHGlobal* X: https://x.com/ETHGlobal​ Warpcast: https://warpcast.com/ethglobal Website: https://ethglobal.com YouTube: https://www.youtube.com/c/ETHGlobal _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _Are you interested in Ethereum development and entrepreneurship?_ 👉 Sign up for the next ETHGlobal event: https://ethglobal.com/events 🎁 Get exclusive access and perks with ETHGlobal Plus! https://ethglobal.com/plus 📣 Want us to throw an event in your city? Tell us where! https://ethglobal.com/city _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

Transcript

[Music] for the next talk I want to bring on eai to tell you about how we get to give allowances to agents so please give eai a big round of [Music] applause hey everyone can you hear me okay okay first that was an awesome presentation from uh we should talk we should integrate uh this concept into Dynamic for recovery so that's actually really cool uh but with that let's let's actually chat about agents with allowances okay so hey everyone I'm II again from Dynamic and today we'll talk about agents with allowances so table of contents uh just like any other presentation I get we'll do a quick intro we'll talk about a secret announcement that we're making today uh we'll talk about some options for AI agents and we'll talk about what's next so very quick overview of why you're stuck with me for the next 19 minutes so I'm the co-founder of a company called Dynamic uh we pretty much provide uh embedded wallet tools so in short really a single SDK for every Wallet transaction or every wallet interaction and those can be very web 3 native and those can be very Web Two uh abstracted experience they can be kind of very in yourf face UI uh kind of interfaces or they can be fully headless and fully abstracted away right so that is what we do in a nutshell you can link multiple doll uh wallets together you can do everything that you ever would wish to do with wallets across evm and God forbid that he said this in an evm conference non evm chains as well but uh that's really what we do we're just context about the company I'm one of the co-founders uh we started in 20 end of 2021 beginning of 2022 we raised about $21 million uh mostly from a16z Founders fund first round capital and others and uh we have a bunch of cool people working uh on the team so just to make sure it's clear Dynamics way more than just embeded wallets we do pretty much anything that has to do with user management right so you log in you can link multiple wallets to the same account you can collect information all that type of stuff we do uh and originally until today we have focused on seamless human to onboarding right so experiences for humans to log in to abstract crypto away from them uh and kind of not think about it all right and you could see this in action today if you go to sites like magic Eden and you log in with either web 3 login or web 2 login that's all powered by Dynamic you can see it on so very web three native experiences you can see it on uh experiences that abstract away crypto so if you go to Arena you log in with Twitter and you're pretty much done you have a wallet you can see this on some upcoming kind of global wallet Partners where you can use a wallet across an entire ecosystem with Incredible Partners like sofon uh and and eclipse and others uh but really Dynamic can be very very flexible right we work with a bunch of folks across the entire industry anywhere from Lamborghini to be chain to Stark net uh it's again live on evm and non- evm chains I know it's an evm conference so I'll I'll you let's pretend that other slide didn't exist but with that let's actually stop Shilling Dynamic for a bit and let's talk agents right so AI agents are going to take over right we'll have these agents perform these tasks for us uh and make our life a little bit uh easier right whether it's to buy things is to wait in queue for us for Taylor Swift tickets which I really hope I wasn't the only one uh or do anything of that sort agents will kind of do it for us so rather than do the task ourself we will tell an agent hey please complete you know please buy a plane ticket for me please plan my vacation and the Agents should have some sort of ability to do that right so it's only natural that we give them wallets not just collect information from us but actually let them spend money on our behalf right an agents will kind of work in two ways in that sense the first one is kind of as an quote unquote agent of us right I will give an agent a task that agent would complete the task they would get some sort of allocation of funds and then they will be able to complete the task pay for it and report back to me and the second type of agent is a very autonomous agent means it wakes up every morning it completes tasks it pays it trades it does whatever and maybe reports back to me so slight Nuance there and we'll see both but the challenge is that you really have to give agents really fine grain permissions right so if think about it if you go to Chad GPT and you write something and it gives you the wrong answer that's fine it's the wrong information mistakes can happen but if you give an agent money and it spends it on the wrong thing or it spends 10x more than you expected that's a really big problem right so permissions and granularity really really matters so we before we chat AI agents we need to chat about how we manage keys for AI agents right how we give an AI agent uh partial permissions to do some things without actually creating uh kind of the ability for them to make mistakes and so at Dynamic we really needed to rethink How We Do key management for agents in order to execute on this right the way we had it set up did not work for this type of use case so today we're actually announcing a full new um Key Management platform and hopefully everyone's excited like me but we'll see uh which is kind of a threshold signature-based MPC uh key management system which allows us to do really fine grain things with giving agents shares to take tasks in defining rule engines for what they can and can't do so we'll talk about that very quickly uh and then we'll talk about how agents use this so very quickly uh who has heard of uh TSS MPC okay who's heard of MPC in general okay great so MPC really means multiparty computation meaning we kind of join together and do some sort of calculation uh in order to get a result TSS MPC so threshold signatures allows us to do this calculation without ever sharing our information so rather than all of us getting a single key and kind of breaking it down each one gets a key share independently and without ever the having the key reconstructed you can create kind of a signature that like a wallet signed right so it's a much more secure method of doing this where someone can get share a someone else can get share B and you can do it without actually uh ever exposing a private key in fact the private key never exists right and so it's really important for a bunch of use cases but one of them is this concept of Agents with permissions so very quickly a quick demo uh of what we do so essentially you can create these really sophisticated flows of two shares three shares four shares uh five shares Etc and decide on the threshold of those how many shares need to be there in order to sign a transaction right do I need to be there in the agent do I need to be there in a server Etc so all of this uh we kind of allow at the kind of very Bare Bones close to the metal infrastructure level on which we can build these more comprehensive Solutions you can do more sophisticated things like re-sharing so I start with I have a wallet I have one share a server has a second share I can reshare and give an agent a two out of three shares so I can start doing really sophisticated things and in order to do everything we talked about we actually have to start with getting the key management right so this is actually coming live in the next couple weeks uh and we'll talk very quickly about the solution so it's very very flexible as a solution uh it's non custodial it can support a myriad of these kind of Key Management uh systems two of two two of three three of five has this built-in rule engine which will come in handy it works out across evm salana Bitcoin Cosmos Etc it leverages tees for Stuff uh and it's really really fast right and that's really important for agents um so just so folks know what that means is we can give these folks limited shares but before that just quickly focusing on how this changes are human focused offering and it's weird to kind of uh talk about human offering and agent offering but here we are uh so we have a few options for wallets for humans right so you're logged in with an email you're logged in with Gmail uh and at that point you can have a two out of two so me the logged in user has a share and dynamic at the server side stores a share you can have more sophisticated stuff with me the end user has two shares and dynamic has one so if I um in a two out of three I can essentially sign things without Dynamic being involved you can have even more sophisticated things and I don't know if it's easy or hard to see but with recovery being completely independent of the dynamic system so these are all examples for a very very flexible key management and wallet management system which is crucial to get AI agents right so that is the thing we started focusing on getting the basics of Key Management right in order to start doing fancy stuff and we'll get to those in a second so just as an FYI it's rolling out in the next couple weeks if you want access to this uh we're we have a booth out there come find us we will talk about it until you get bored but uh come find us but very quickly now moving to AI agents now we can talk about this right so not sure if you heard AI agents are the rage right they do some weird things like meme trading and Bot trading and things like that but they also are useful for very day-to-day activities again book a trip uh monitor things on my behalf right take action on my behalf renew a license on my behalf all these types of in experiences that we manually do today that technically we should be able to delegate and give agents money information complete a task on our behalf right and so um there are lots of cool Frameworks out there I'm not going to focus on them I just want to focus on the wallet part right so we talked about this there really two types of um ways to work with agents right semi-autonomous and fully autonomous so let's start with human and the loop semi-autonomous so these are the use cases of trading stocks uh booking flights Etc you need this rule engine oversight uh and it has some sort of control autonomy So within a confined set of rules it can do some stuff and the way that works is the following so let's walk through this example so a user goes to an AI travel website right a travel agent website they Define some sort of parameters on the trip as well as a budget or I as a user Define that the agent executes on those the policy engine sits on the dynamic share on the server share and ensures that signing cannot happen without that server share so agent gets and we'll see this in the next screen an agent gets delegated some sort of user share but they cannot sign without the agent share right so you can start doing this with wallets but technically you should be also be able to do this with identity right so I logged in me as a human have delegated my passport information similar to the sell presentation that that uh was just here which was really really cool and relevant uh I have delegated some sort of information to the agent they can act upon it on my behalf and then report back to me with the results right and me at the end of this I go on vacation and I think about how weird this place is with AI agents and human at the same time right so the way that works is I as a user get a share right MPC model means shares right I as a user gets the share there's a server share and I can generate turn from a two of two to two of three and generate an additional share to give to an agent then can bind some sort of policy permissions on what that can do right so again I essentially take if I had a wallet take the wallet give a credit card but Define what's in that credit card and what can happen or issue a virtual credit card for all intents and purposes the agent takes care of what it is they want to take care of and kind of reports back to me with the results right so that is a human in the loop model there's also a second model which is no human in the loop right in this model the entire concept of interfaces kind of goes away right the idea is that an agent wakes up in the morning and can just accomplish tasks and the flow is the following so an agent wakes up in the morning they have their coffee uh and then they kind of go in and server only serers side only kind of get some sort of key from the server with permissions they execute on everything with no interfaces just run on top of things execute and report back with the result again the end of this is I go on vacation and uh I think about the weirdness of this world but essentially in this scenario we have server full server side wallets very low level serers side wallets they allow you to do this without any interfaces at all right the architecture of this is an AI agent wakes up in the morning asks the dynamic system for a temporary wallet the dynamic system issues that again it issues it doesn't send a share but sends the logic to create a share for the agent on their end right they actually uh do the calculation sign whatever messages they want to sign within the confines of rule engines and policy engines and they can move forward right so in this instance you actually have an entire no human in the loop type interaction so again two types of things no human in the loop where it's just server to server machine to machine type interactions with wallets or human in the loop where a human has some sort of interface delegation and then end of Delegation so we likely see a combination of both of right so in both of those ideas the following which is pretty straightforward which is we will do less work right the idea is that we need to do less and less work we get to relax a little bit more and again reflect on the weirdness of Agents doing stuff on our behalf right uh but one more thing and one more really cool thing that will happen next couple weeks is you can start expanding this with who's heard of 7702 okay perfect so uh for those that don't know uh and this is a Shameless plug around kind of an ethereum uh thing that's rolling out in the next couple weeks ethereum is rolling out the Petra update uh that's going to have 7702 support then inherently means that you can delegate an eoa account or a wallet to some sort of smart contract so you can do things like gasless so if you log in with metamask you can still do gasless type interactions right or a coinbase etc this is the next part of that solution so an agent interacts on our behalf can we also abstract away gas from them right and something like 7702 that will come in will let us do that so with that let's summarize what we talked about so we gave an overview of why AI agine Circle right we talked about this new TSS MPC solution and again AI agents are a part of it you can use it multiple types of solutions it's really really cool want to see is that did that work yeah okay cool cool cool checking yeah okay whoa okay W if you want to see a demo of it we're around we're happy show you oh it's me I'm I'm the problem it's me reference yeah yeah no who got that no oh my God not enough people thank you I appreciate that okay it's a Taylor Swift reference come on um okay so we talked about this TSS MPC offering it can do a bunch of things um whomever is a Taylor Swift fan can see a demo otherwise don't talk to me uh but besides that we have these two paths uh for using these agents and we talked about humans doing less things so uh I will end with the following we're in the business of giving people tools to create these innovative ideas right so I gave some examples of using AI agents for travel for trading Etc but the goal for us is actually to provide low-level tools that let people come up with ideas we did not think about right if if next year's presentation I present and I show all the cool travel agent tools that have been developed I'll be really sad right so my ask kind of leaving you with this is my ask is to actually rethink a lot of how agents can be used and come chat with us about um different ways to actually use AI agents in ways we haven't thought about right and our goal at Dynamic is to give you tools to do this in a way that's very very flexible right surprise us with some of these ideas we are going to try to get out of your way and give you more powerful and secure tools to manage money and to manage kind of very sensitive identity type interactions while you the collective you think about some of the cool ideas for next year two and five and with that thank you [Music]

Automatic transcript — names and jargon may be misspelled.