# Agentic Commerce: Infrastructure for Autonomous Agent Payments | Rick (March 2026)

- Channel: [Berlin Ethereum Meetup](https://streameth.org/berlin-ethereum-meetup)
- Date: 2026-05-09
- Duration: 26:29
- Watch: https://streameth.org/watch/yt-fcKvD358Qmw
- YouTube: https://www.youtube.com/watch?v=fcKvD358Qmw

## Description

Join us on Meetup to get up to date on our Berlin events: 
https://www.meetup.com/berlin-ethereu...
See you at the next one!

Apply to speak at our future meetups: https://forms.gle/txTvFB4E8E8KbEqX8

X (Twitter): @BerlinMeetup
https://x.com/BerlinMeetup

## Transcript

I'm Rick. I'm with the developer acceleration team at DF. Um I'm mainly focused on helping builders um figure out what tools they need when they come when it comes to building in web 3. Nowadays that changed a lot. So today it's like understanding how we can help in the agentic engineering flow to make sure that it uses the right tools in web 3. So we have created skill files. Esskills.com is for example one uh and many other tools to make that possible. And today I'm going to talk a bit about agentic commerce. Um it's a very interesting um new thing that I think will spawn out uh exciting opportunities especially to also increase adoption of stable coins and cryptocurrencies overall. Um and while in this talk I'm mainly going to focus on the entire stack and everything that Ethereum foundation has built especially with the DEI team. I also want to mention a bit why this space is exciting or interesting and why you should care. So whenever it comes to introducing a new payment shift, it was always about removing a layer of friction. Yeah. So you see in the uh 60s we started with Visa Masac which basically offered the opportunities to just pay uh anywhere. Stripe ran when we had more and more payment providers solved the problem to basically aggregate all payment providers for e-commerce shop together. So you only need to integrate Stripe and not need to integrate 10 to 15 different payment providers. And X42, the agentic commerce layer with crypto, is basically solving another friction point, which is not on the merchant side, not a developer that actually creates a shop and wants to sell his things via payment methods. It's actually the consumer side. What does this consumer side look like? Um, I'm going to make a quick example here. So, let's say everyone is a builder now and has it their agents. um they come up with a bunch of ideas and code writing code right now is very easy because the agent does everything for you. The part that's still manual is when you have to access services like APIs for example to um build stuff on top of them. Yeah. So let's say for example I want to build something where I'm going to use midjourney to create an image. I want to use an AI model that takes these images together, stitches them into a video. Then I want to use an enrichment API to find customers that I can send this video to. and use an email platform to send it. Now, this could be like just anything really. It can be also if I just want to hand an analytics dashboard for my website, any service where I basically need to create an API key for. And the friction here is that I have to create an account, I have to add a payment method, and then I have to generate an API key. Yeah, this is the part that creates friction. And if I'm sitting right now on my couch and I'm using like Telegram to code, which is for me a common occurrence in the last couple months, um I'm bound by my laziness to not do these kind of steps and not execute on ideas where I need API keys to solve this. Yeah. And this is where Gentcommerce, especially in the stable coin world, um it removes this friction because you basically don't need an API key anymore. And the way how it works is, and I created a quick overview here, I'm going to use agent cache as an example. Um, you can just set this up. It creates a wallet for you. Um, they even prefund it with some USDC. Your agent now, let's say this is an open claw harness, has access to this wallet with some stable coins funded in it. And then it can basically find all kinds of services that you can basically um pay money for to use them on demand. So, you don't need an API key. You don't need to read documentation on which API um endpoint you need to hit. You can basically just look at all the different services that are available and then pay on demand for that. Yeah. And these are send microp payments. Yeah. This is where stable coins come into place and where it's very important to have minimal fees. Yeah. And this also is very important that we have layers and infrastructure that basically is scalable and also microp payments are going to be a lot. Yeah. We're going to not run in just a few transactions per day. These can be thousands of transactions uh just from a few agents. So we need a very very low um fee structure. Um so this is the basic flow that agents now have access to. This works in production right now and this has this advantage of not having an API key but also it has these features that it's permissionless the fees and it's open source. It builds primitives that everyone can build on top of. Yeah. Which is awesome because now it opens up opportunities for bu people to build application layers on top of them. Yeah. And obviously this this permissionless aspect is also interesting because Stripe has u frozen accounts from a lot of SAS people especially indie hacker for multiple reasons. And with crypto or with stable coins you're basically not bound by these platforms anymore. So on the one side we have this new primitive on the left side. This is kind of like uh another a use case for gentic commerce. This is where things get interesting where steps are just cumbersome. So if you for example want to book your flight I booked my flight three four weeks ago with openclaw to San Francisco. Everything just with openclaw operating the website and so on. It's amazing. It's cool. Um also like every year you want to change maybe your utilities providers to save money. I'm too lazy. I'm not going to spend a few hours to research what utility provider I should use, but my agent can't do and saves me a couple hundred bucks. The left side here, this is where virtual credit card still makes sense. Now, this is where Misa, Mastercard, everyone is tackling this side now to get your agents to use them to pay for things and to handle research and execute that. The right side, that's where the stable coins and the crypto side makes sense. Yeah. for the microp payment use case and for basically other use cases where agents want to hire themselves and not themselves but hire other agents. So this is a little bit of reasoning. Now let's get into the entire stack. There are a lot of standards built in here. Um X42 is everything that kick the the standard that kickstarted everything the payment layer. We have EOC 8004 which is the trust and discovery layer. Then we have 8183 which is basically the commerce layer. It wraps the payment and the trust and discovery together in case you need additional features. Then we have 8128 which is the authentication layer. And then we have also privacy. Yeah, this is also becomes relevant um especially if you have uh transactions that um make it possible for people to link payments and to map customer behavior. So let's talk first about 8004. Um a lot of people when they think about a4 this is how got kind of like misinterpreted think it's for your agent to create an identity. So you want to give your open call an identity use 8004 and now it lives on the chain. This is actually a use case I would say that doesn't really make a lot of sense and the most important one is because again privacy why would you expose your agent to to a chain and give yourself an attack vector where your agent is explorable where 8004 actually makes sense is for service providers. Yeah. So this is actually made and you can read this in the spec for MCP servers, for APIs that offer just any kind of like service like I mentioned before, for example, inference endpoints, for really any kind of service that wants to sell something um to you or to your agent. And the 8004 specs consists of three different registries. On the one side, the identity layer, then the reputation and the validation. And um I'll quickly just go through them one by one. The identity itself is quite simple. It's just an NFT that you're basically minting. It consists of all the capabilities you um you can do and reputation will be stored in it and the NFT is also portable. So in case you want to youu uh reuse the the identity on a different wallet, you can just send it over. The reputation layer is the most interesting one probably because it basically is bound by the transaction. So if you want to make a transaction to to an agent, you want to pay for a service and you want to leave a review later, you need to prove that you paid for this service. This is obviously on one side for civil resistance, but on the other side also in general just cool because now I can actually prove that the people that make reviews actually used my service. There's still an attack vector here. Obviously, I can just create a thousand agents, fund them with an agentic wallet, and then basically pay myself. Um, I still have to pay the fee on the L2, but uh this is still possible. However, the defense layer is that you can trace these funds. You can tag wallets that only pay uh once your own angel and then there's also the flagging part which you can trace circular payment patterns. Yeah. So this is not exactly solved by the spec. There's not an application layer. The spec is basically building the entire um trace of the transactions and the reputation. But um basically application layers then can aggregate the spec and can do all these things and and create websites where they can explore them. The validation registry is the last one and I think it's a very interesting one. And to make an example here, let's say for example, I'm selling inference and I'm claiming that I'm selling you GPOSS 120B. Maybe it's wrapped around an agentic swarm service. And now, how can you actually prove that you as a service provider are actually offering 120B and not GP2s 20B? Yeah, it's very hard to prove that just by the result because sometimes the prompts are just so easy and uh 20B actually is indistinguishable from 120B. This is where TE antations come into place. So here you can actually prove that you're running the code on the server um that you're claiming. ZKML proofs are also available and there's also the staking aspect where basically the validator that takes from the service provider the cryptographic proofs and actually checks them and then stores them in the validation registry can stake to make sure that if they actually handle in any malicious way they're getting their stake slashed. So this is a cool way to secure this entire system. So this is the entire 8004 spec. I think it's the most exciting one. The X42 is obviously the most um basic one which is just a payment layer. It's based on HTTP 402 status code which was dormant for the last decades and now basically Coinbase started to introduce this as a payment layer finally and it's very simple. Basically you as a client you can send an HTTP request to a server. The server responds with the status code and says the payment is required. Then you perform that payment and you sign it and after the payment is live on the blockchain you can just send the payment signature you sign it as well and the server verifies it and returns the result. Yeah, this is basically when you want to do upfront payments where there's no additional agreement whatsoever. The one part that is missing X42 is actually the authentication itself. So how can I sign the HTTP HTTP request? Obviously, if I'm making a payment, I want to make sure that other people cannot claim that they paid for the service and get the result themselves. So, this is where every issue request needs to be signed with my wallet. And this is how the service provider can ensure that the person that is requesting the result actually paid for it. Now, we went through the discovery side, the trust side, payment and authentication. Um the one thing that combines a lot of these standards together is 81 83. This is what I call the commerce layer. So in here basically we can add additional features if an upfront payment is not the only requirement. So for example, think of a use case like Fiverr or Upwork or you have escro payments where you want to not pay upfront but you want to prove that you're willing to pay and where there's also a dispute resolution section. So if you escort money and the service provider contributed um the result um and you as a customer claims that the result is not um not good, you can dispute it and then it can be a third party or an AI or CK verifier that disputes um via an escrow. So this the job that basically Fiverr and Upwork did as a plat a platform now is fully permissionless in year and and this is awesome because I actually in 2022 created a startup where we created decentralized marketplace for uh developers and we built this entire stack on our own and there were a bunch of other startups that built this entire stack themselves as well and then everyone had their own reputation layer. Everyone had their own escro protocol. Everyone build basically the same thing and then people try to aggregate everything and had to use different standards. Now this is standardized. Yeah. So everyone can just rely on this and build applications and when it comes to reputation, when it comes to handling disputes, it's just the same way on the standard side. So we went now through the entire flow how it looks like from the agent to requesting a service and then paying for it. The one thing I didn't mention is when we're removing API keys, there's just one last key that remains which is the private key of your wallet. And the question here is how do you actually store it? Because obviously you don't want to store your private key in plain text right now. If you're giving your agent access to your codebase, even if you add in the agents MD file that please don't look at the environment variable, it will still do. Yeah, there's actually no way to get LMS especially like cloud code type of tools um and enforced instruction to not read environment variables. Yeah. So understanding what is the best way to handle private keys for your wallets um is actually becoming very important and there are different type of solutions that come to mind encrypted key stores hardware wallets and even cloud secret managers but in the end all of these are never 100% perfect and so there's an open wallet standard from moon actually released I think last week which is trying to solve this on the one side we have other emerging frameworks works like vlog.deaf that is not focused on wallets especially it's more focused on how can I handle my environment available in a safe way so that they cannot read it. Um but overall I think this the the future this is what everyone is building right now. All the wallet providers are trying to build their own agentic wallet solution right now. The solution are basically session keys where you can give spending limits to your wallet where you can spawn wallets for specific use cases where if you say hey please do a trade on unis swap you can basically just give your session key uh the permission to just perform swaps on on unis swap and then just make it not usable again after 24 hours. Yeah. So your keys rotate um every every other day and um it has very very limit uh limited exposing. And the last part which is not specifically just built for this entire gent space is the privacy layer. So stealth addresses, you probably heard about them. Um I think this is very cool because it on one side allows the consumer to stay private. I don't want other people to know what I'm buying. But then also on the merchant side, it's interesting because competitors could basically map consumer spending patterns and understand through this how consumers handle this service. They could make uh research. They could basically get competitive advantage through it. And privacy basically makes it possible to um completely negate that. And it works without mixers. It's a basically a very very good uh way to u do this. um without breaking composibility. Yeah. And this is basically the entire flow here again showcasing it uh in this overview. Um we have the payment layer, the authentication layer, the trust and discovery layer and then afterwards we have the entire privacy wrapped around it and if you need additional features the asone settlement layer via 8183 comes into place. Yeah, so far this is what is already in production. I see tons of use cases here with upfront payments. The escro side is still very very new. So this is what people are building on top right now. And especially if you're looking for opportunities, I think this is where most new things still have uh possibility to flourish um in an environment where you don't have a lot of competitors yet. So now we went through all the different standards. people already built entire frameworks to actually access these standards to make it very easy to implement them. I created an entire map where you can see how every single um basically category uh is looking like right now, what frameworks they're out, what tools they are out, what companies are building. I also made it clickable so you can click on the different tools if they're not specific related to Aentic commerce and there's a link to a blog where they're mentioning how their tool is relevant in the Aentic commerce space and you can basically from there figure out what tools you need to um make your agent um basically engage in the agent commerce space. One more thing that I would like to kind of like talk about this is just like some some last I would say thing to wrap this uh talk up. I think an exciting opportunity for the entire agent commerce space is obviously obviously to increase crypto adoption because if you think back about the slide where I said what are the fiat the virtual credit card use cases and what are the agentic commerce use cases for crypto. I think um there are a lot of connections and the cool thing is if now agents have access to virtual credit cards, it's actually very easy for the agent to also use these credit cards to on-ramp into stable coins. And now suddenly people are automatically onboarded into crypto and the entire UX program that crypto had for the last decade is basically removed. Yeah. Because agents handle everything for you and they can also start doing payments in DeFi. And once basically merchants see an very big spike where now we're sitting in I think X42 has a volume of 40 million payments so far. Once we see merchants actually getting more excited about like using stable coins and accepting stable coins and this goes up more and more and more people will have access to agents with virtual credit cards. You will see merchants actually also incentivizing agents to pay in stable cards stable coins instead of virtual credit cards with like Visa Mastercard. And this is where I think this obviously this is all an assumption but this is what makes me exciting about the space. I think this is what can kickstart a huge wave for the adoption of crypto. And I think yeah this is how I would like to wrap up this um this talk. If you have any further questions on any of the different standards feel free to hit me up. Um the QR code also actually on the website uh if you go there has my contact details so you can write me an email. um or you can get my telegram there and if you want to build just play around with the frameworks and ask your um open claw to use skill files to interact with it. Thanks. >> Are there any questions? >> Yep. I just want to double confirm. First of all, thank you. I think this is very promising. Um you said the first three steps in the slides are in production and the settlement is in discovery mode right now. I just want to double confirm what is usable right now. >> Um everything is usable right now. The entire ASCOR and settlement aspect I think just has a lot of opportunities for the applications now to use this and and build like frameworks around it. Um in production everything in in this instant settlement you can just play around with it. You can like basically pay for any API requests right now. um everything that has like um the intention to actually set up escros and stuff like that. This is I think still underexplored because the standards is very new. So um you can build a framework around it now. Yeah. >> Yes. >> Um I feel like for me personally, I wouldn't trust my agent to make good decisions when buying things. Uh because I don't know like I think there's still some element to it where I'm like oh will the agent actually buy good value products or something? I don't know. Um so I think what is what is like stopping agents from not just going out there with your stable coin wallets uh blowing all the budget on like some random APIs or like some endpoints. um like how does the agent know if it's quality or like worth spending money on certain endpoints versus not? >> Yeah, it's a good question and the interesting part is that what Visa Mascot is trying to tell you is that the agents will know things that you need to buy before you even know it. I don't buy that vision. I think generally the idea is for us to just remove friction. So in the end, I mean, if you give your agent's wallet a specific amount of permissions what it's allowed to buy, that's how you can circumvent it's buying stuff that is not relevant to you. Yeah. And then also, I think there should be always a flow where the agent asks actually like, hey, should I really buy this? Obviously, if you do dangerously skip permissions with your cloud code, then in some cases, you can get wrecked and it just buys a bunch of things for you. But that's where also spending limits make a lot of sense. Yeah. So I think it's mostly how the wallet is managed with spending limits and giving it permissions to buy certain things and and other things you just don't give it permissions for and the wallets also expiring after a day or two and you need to delegate it again uh a new wallet to to continue. >> Yes. >> What are the use cases you are you are excited about that that haven't been possible so far that you've seen growing now? >> Yeah. Um, I think it's it's really just a friction part. So, for example, when I created the the dashboard, what you saw there, I added analytics to it and I didn't want to set up an API key and I just basically told my agent with his wallet like, hey, please just get an API key for the service. Uh, it found on a in a skill file a service that accepts um microp payments via X42 and it just g it just generated the analytics API key via that. Yeah. So I think there's these these use cases where right now we can just natively onboard into tools that are have this um this gate. Yeah. Where I need to do things um to actually open it and I think that's very exciting. But then also like when I don't know if you have like guys how many of you guys have an open claw or like an Hermes agent or whatever, right? But if you do this he he thought it was a question. No, but we can ask it like how many of you guys actually have an open claw right now? Okay, perfect. So, I think this is also very exciting because I have tons of ideas, but again, I'm limited and and bound by my laziness. Like, if I want to let the agent create an image for a website and I don't want to set up a midjourney API key so that it can actually do these things, I'm just not going to do it. And a lot of ideas will just sit on a shelf and not being executed. And also another thing which is I think very important is if I'm using midjourney I'm actually um bound by their tier system oftent times sometimes they have fixed credits but sometimes it's a fixed tier system. So I'm actually sometimes forgetting to unsubscribe for a service they just need once then continue to pay for it and also often the the tier system is just way too high and I'm overspending. So just paying on demand makes so much sense. And I think yeah, removing the friction and and and basically killing my uh my barriers and setting myself because I'm lazy. I think that's what makes it most exciting. >> How do you like envision because I think that if that would be the future, but like all the businesses who have their model based on subscription where you just don't cancel it. So you're paying for Netflix for next two years. uh like could kind of change probably if that would be the future which really happens what do you kind of envision there like what would be kind of winners losers um >> you mean what the entire agent commerce space maybe going to exclude in the market >> like some things we can imagine like it's a personal system just does things for me I think that's what we can imagine there's quite a lot of things maybe it's hard to kind of like get from my point of view because it's like something new use cases which are kind of not really That's correct. Didn't exist there before. >> Yeah. I think everyone is making speculations right now and nobody knows in the end. Um I think what a lot of people are criticizing is that AI is not going to or should not do emotional buys. So I think the the services that actually offer like hotels and and other things who actually care about like hey what does does this look like or furniture or art like all these things I think you can exclude that that will not be part of. Um, but I'm I'm not sure if I understood your question right. I think in general the the losers in the end, there's no loser. Yeah, you're just trying to amplify what already exists and you're removing the friction. I think maybe the losers could be those that don't add um, you know, to their data API and and and endpoint for X42. I think those are the losers, the people that don't see that this has right now great opportunities to remove friction. Thank you. >> Thanks. [applause]