Agent Spending Without Controls — Rodrigo Coelho & Pranav Maheshwari, Edge & Node
Description
Pranav Maheshwari ran one prompt in two terminals: find the email address of the head of crypto and blockchain at Mastercard. The terminal without a payment skill file returned the company's email format and an invitation to work the rest out. The one with it paid a fraction of a cent to a metered endpoint and came back with the address, the person's location and their handle. That gap is the whole argument. Most MCP servers are free today, and his claim is that the useful ones will not stay free, which leaves an agent only as capable as the tools it can pay for. Rodrigo Coelho sets up the longer history first: Edge & Node built The Graph, served 1.8 trillion onchain queries, and shipped a query micropayment system in 2021 that cited the HTTP 402 spec years before Coinbase released x402. What stops enterprises adopting any of this, in Coelho's telling, is compliance rather than throughput. Traditional rails assume a human somewhere in the decision loop. Agents transact around the clock at machine speed, and a counterparty arrives as a bare wallet address with no identity attached to it. Somebody carrying a title like chief legal officer has to sign off, knowing that getting sanctions screening wrong carries fines in the billions. The closing demo makes that concrete. Two agents hit a metered scraping service, one from an ordinary wallet and one from a wallet flagged as sanctioned. With screening off, both transactions authorize. Switching screening on leaves the first working and rejects the second as a blocklisted address. Speaker info: - https://x.com/rodventures - https://www.linkedin.com/in/rodrigoco/ - https://x.com/impranavm_ - https://www.linkedin.com/in/thepranavmaheshwari/ Timestamps: 0:00 - The Graph, and paying for queries before x402 existed 2:03 - Prior art: micropayments, the 402 spec, and joining the foundation 3:52 - Rails built for humans in the loop, not machines 4:49 - Why enterprises stall without a compliance layer 5:42 - A wallet address wit
Summary
Generated by gpt-5.6-terraAt-a-Glance
- Verdict: Skim
- Core thesis: Agent commerce will require a machine-native payment and compliance layer because paid MCP/API tools, autonomous purchasing, and micropayments cannot safely rely on human-oriented checkout and card controls.
- Why it matters: As agents gain authority to call paid tools and purchase goods, the key bottleneck shifts from model capability to delegated spending limits, transaction traceability, counterparty screening, and policy enforcement.
- Best use: Use this as a concise market and architecture signal for agent-payment infrastructure, especially the need to put compliance screening directly in the transaction authorization path; do not treat the product demos as proof of production readiness.
Executive Summary
Edge & Node presents Ampersand as a financial harness for agentic commerce. Its central argument is that agents will increasingly need to pay for MCP tools, APIs, web data, and retail purchases, but traditional payment rails assume a human makes or approves each decision. A wallet address and an autonomous agent do not provide the identity, controls, auditability, or compliance assurances an enterprise merchant or legal team needs.
The product thesis is an aggregator model: install an Ampersand skill file into an agent environment, let the agent access paid MCP endpoints, and settle the underlying microtransactions through a wallet rather than requiring a separate account and credit card for every tool provider. The demo contrasts an agent without the paid-tool integration, which can infer a Mastercard executive's email pattern but not retrieve the exact information, with an Ampersand-enabled agent that reportedly pays for and returns richer contact data.
A second demo shows terminal-driven Shopify Universal Commerce Protocol purchasing: the agent searches for a Father's Day gift under $10, uses stored user data rather than re-entering checkout details, and places an order through the Ampersand wallet. The more consequential demo is compliance enforcement: a simulated sanctioned 'bad claw' wallet can make a 0.1-cent scraping-data payment when screening is disabled, but is rejected after Ampersand enables wallet screening through TRM.
The useful strategic point is not that agent spending is immediately enterprise-ready—the speakers explicitly say enterprise deployment remains early and largely experimental—but that payment authorization for agents needs to become a programmable policy-control plane. The talk is product-forward and light on implementation specifics such as scoped credentials, approval thresholds, budget policies, dispute handling, or non-blockchain identity, so it is best skimmed for its framing and compliance pattern.
Key Takeaways
- Claim: Paid MCP tools and APIs are likely to become a primary way agents gain differentiated capabilities, creating a need for low-friction machine-to-machine payment. | Evidence: The speakers argue that tools such as Exa and Firecrawl are currently often accessed through individual accounts and credit cards, but demonstrate an agent using an installed Ampersand skill file to invoke a paid data endpoint and generate a transaction in the background. | Implication: Treat payment access as part of agent tool provisioning: the control plane should know which tools an agent may buy from, at what price, and under which budget and policy. | Caveat: This is a directional market claim rather than evidence that paid MCP usage has already become the dominant agent pattern; the speakers characterize the market as early and experimental.
- Claim: Human-designed payment flows are structurally inadequate for agents that transact continuously and at machine speed. | Evidence: Rodrigo Coelho contrasts human-in-the-loop financial approval with agents operating around the clock and at microsecond speed, arguing that existing rules and controls were not designed for entities that autonomously initiate many transactions. | Implication: Do not give production agents broad card access and assume existing payment controls are sufficient; authorization must be evaluated and enforced programmatically before each transaction. | Caveat: The talk establishes the mismatch conceptually but does not quantify the latency, fraud, or operations failures of existing card and checkout rails.
- Claim: Compliance screening is a prerequisite for merchant and enterprise adoption of agent payments, not an optional post-transaction monitoring feature. | Evidence: The presenters cite sanctions, terrorist-financing exposure, counterparty identity, and fines that can reach tens of millions to billions of dollars. Their demo uses TRM screening to identify a simulated sanctioned wallet and deny its attempted payment. | Implication: For any agent-payment stack, make sanctions/KYT screening, counterparty policy, transaction logs, and deny decisions first-class authorization controls on both buyer and merchant sides. | Caveat: The 'bad claw' sanctioned-wallet scenario is explicitly a simulation, and the presentation does not address how screening resolves attribution, false positives, or a legitimate user operating through a new or privacy-preserving wallet.
- Claim: An agent payment wallet can remove repetitive checkout friction by combining stored user context, tool access, and transaction settlement. | Evidence: In the Shopify UCP demo, the agent receives the instruction to buy a Father's Day gift below $10, finds options, reportedly uses remembered name/address/phone information, completes an order, and displays a receipt and tracking information without manual checkout entry. | Implication: Convenient autonomous checkout increases the need for explicit delegation policies: per-merchant allowlists, item/category restrictions, price ceilings, approval triggers, and constrained use of personal data. | Caveat: The demo does not show consent boundaries for remembered personal data, spending approval before the final purchase, refund/dispute handling, or protections against prompt-driven purchasing mistakes.
- Claim: Micropayments may provide an alternative web monetization path when agents consume content directly and traditional ad-supported browsing loses relevance. | Evidence: The speakers point to Cloudflare opening a gateway through x402 and propose that an agent crawling a site could pay a microtransaction for the information it needs rather than viewing ads; their compliance demo uses a 0.1-cent payment for a scraping service. | Implication: For data-agent products, monitor x402-style paid-access models as a possible procurement mechanism, but preserve fallbacks for subscription, enterprise contract, and non-crypto billing arrangements. | Caveat: No evidence is offered that publishers, users, or web platforms will broadly prefer per-request micropayments over subscriptions, licensing, ad models, or bot blocking.
- Claim: Ampersand is positioning itself as infrastructure at the intersection of agent wallets, paid tool aggregation, and compliance controls. | Evidence: Edge & Node says it incubated Ampersand from prior Graph Protocol work, including a 2021 query-micropayment system; it later collaborated with Coinbase and Google on x402 and says it contributed batching concepts intended to reduce gas costs for nano-payments. | Implication: Evaluate Ampersand as one emerging infrastructure vendor, but separate its valid control-plane problem statement from unverified claims about product maturity and market inevitability. | Caveat: The presentation is a vendor talk and does not provide independent adoption data, pricing, security architecture, reliability metrics, or a detailed comparison with other agent-payment providers.
Detailed Brief
Market timing and protocol context
- Claims: The speakers frame agentic commerce as a transition comparable to the multi-decade move from early electronic transactions to online e-commerce.; They say Edge & Node began exploring agent payments for Graph Protocol data-query use cases in late 2024, shortly before Coinbase released x402.; They describe the ecosystem as expanding across payment protocols, governance layers, and infrastructure providers, while enterprise rollout is still not broadly underway.
- Evidence: Edge & Node says the Graph Protocol has served 1.8 trillion on-chain-data queries since 2018.; The company says it referenced HTTP 402 in a 2021 micropayments blog post and later joined the x402 Foundation.; Circle is named as having introduced a related 'nano payments' approach.
- Caveats: The claimed prior art and protocol contributions are presented by the vendor and are not independently substantiated in the transcript.; Transaction-growth references for x402 are asserted without a visible number, time period, or source in the spoken transcript.
- Implications: The relevant near-term opportunity is likely infrastructure for paid data/tool access rather than immediate mass consumer agent checkout.; Protocol positioning may matter, but a production architecture should avoid being dependent on a single wallet, chain, or payment-standard assumption.
Control gaps the presentation leaves unresolved
- Claims: The talk correctly emphasizes sanctions screening but defines the broader governance problem more widely: preventing hallucinated actions, overspending, and policy breaches.; Its demos focus on transaction success and sanctions denial rather than the full lifecycle of delegated financial authority.
- Evidence: Rodrigo says a chief legal officer or chief policy officer must be confident that agents cannot hallucinate, go off rails, overspend, or break policy.; The demonstrated control is a binary screening toggle that changes a flagged wallet's transaction from authorized to rejected.
- Caveats: There is no demonstrated mechanism for recurring budget enforcement, vendor or SKU restrictions, multi-step approvals, human escalation, revocation, receipt reconciliation, chargebacks, fraud recovery, or policy testing.; The transcript does not explain whether agent identity, wallet ownership, customer identity, and authority delegation are cryptographically or contractually linked.
- Implications: A sanctions check alone should not qualify an agent-payment system as enterprise governed.; Any evaluation should require a decision log that records the requesting agent, delegated authority, policy version, vendor, amount, data accessed, screening result, approval path, and settlement outcome.
Notable Concepts & Terms
- Ampersand: Edge & Node's incubated product for agentic commerce, presented as a wallet/payment layer that gives agents access to paid tools and enforces compliance checks.
- x402: A Coinbase-associated payment protocol/specification invoked as infrastructure for machine-to-machine web and API payments, particularly micropayments.
- MCP servers: Model Context Protocol tool endpoints; the talk's core commercial assumption is that valuable agent tools will increasingly charge for usage.
- Skill file: The installation artifact used in the demo to give an agent access to Ampersand's tool/payment marketplace rather than configuring each paid MCP provider individually.
- Shopify UCP: Shopify Universal Commerce Protocol, used in the demo as a pathway for an agent to search for and purchase a retail item from the terminal.
- TRM: The compliance-screening provider integrated into the demo to inspect a wallet and reject a transaction from a simulated blocklisted address.
- Agentic checkout / financial harness: The presenters' framing for autonomous purchasing constrained by payment, identity, compliance, and policy controls rather than ordinary consumer checkout flows.
Operator Notes / Why Ken Should Care
- Define a delegated-spend policy model before enabling paid tools: agent identity, permitted vendors/tools, per-call and rolling budgets, categories, approval thresholds, and emergency revocation.
- Require pre-authorization controls beyond sanctions screening, including prompt-injection-resistant purchase intent, merchant allowlists, price/quantity anomaly detection, and human approval for irreversible or personal-data-bearing purchases.
- Instrument an immutable transaction-and-decision ledger linking each payment to the agent run, user authorization, policy version, MCP request, counterparty screening result, receipt, and reconciliation status.
- Test x402/MCP payment rails as an experimental procurement path for paid data services, but avoid coupling core workflows to crypto wallets or a single protocol until identity, accounting, refund, and legal controls are validated.
- During vendor diligence on Ampersand or equivalents, ask for evidence on custody/key management, credential scoping, policy engine semantics, spend-limit enforcement, screening coverage and false-positive handling, dispute support, and enterprise audit exports.
Source/Metadata
- Title: Agent Spending Without Controls — Rodrigo Coelho & Pranav Maheshwari, Edge & Node
- Transcript words: 3480
- Duration seconds: 1248
- Timestamp note: No timestamps or chapters were present. The transcript repeats the latter portion of the payment and compliance demo.
Transcript
All right, thanks for having me. Today we're going to be speaking about ampersand and agent spending without controls, which is the missing infrastructure layer for AI payments. A little bit about myself. I'm CEO of Edge and Node. For those that aren't aware, we were the team that built the graph protocol. For those that aren't aware of that, it is a blockchain data indexing protocol. It's been around since 2018. We have a decentralized network. We served 1.8 trillion queries over the years of on-chain data to applications utilizing blockchain data. I've been a serial entrepreneur myself. We've incubated a project within Edge and Node called ampersand in the ejected commerce space, which is what we're going to be talking about today. And so we're all aware that payments have been online for the past 30 years. Starting in 1981, we had our first electronic transaction. And fast forward, we had a multi-decade transition. And this is from credit card payments to online e-commerce payments. Fast forwarding all the way to today, where we have the rise of the aging economy and unlocking a new wave of economic value. And now we have LLMs and agents that can transact. And we have this underlying payment infrastructure that's been around for decades. And it's being rebuilt today for the agentic economy. We've heard from many speakers here about how that's being done. We're going to be discussing today our perspective and angle on it. So we ourselves at Edge and Node were really early into the agentic commerce world. We actually developed a micropayment system for queries back in 2021. We already referenced the 402 spec back then in a blog post. And so back in late 2024, we were looking into agents and how they would interact with data, how they would pay for it, specifically for the graph protocol itself and making micropayments for queries. And in our research, we were looking at how this could happen. And then a few weeks later, X402 was released by Coinbase. So we jumped on immediately with the Coinbase team, with the Google team, started to collaborate on the X402 spec. We've joined the foundation. We've contributed to the X402 spec ourselves, leveraging some of our prior art on the micropayment system and batching protocol as part of how it could be utilized to reduce gas fees when you're dealing in nano payments. Circle themselves came out with their own version of it called nano payments. So we were also working in similar veins. And so we've seen an explosion of companies emerge, solutions, technologies, and this agentic commerce map. And so we're just one little piece in here of people developing infrastructure, payment protocols, governance, etc. And so traditional payment rails, though, were built for humans. When we're dealing with financial institutions specifically, there's a human in the loop that sets the decision-making process of whether that payment is allowed to go through. But we're dealing with agents that transact at machine speed and around the clock. And all these controls and policies and rules were built for humans and not for machines that don't breathe. Right? So we're dealing in microsecond machine speed, and it's just simply not going to work in the way it has. And so we need a new method of doing that. You can see a lot of the transactions exploding. This is over time. We're seeing transaction growth over X4O2. We're all saying this is coming. It is still early days. We ourselves are seeing a lot of experimentation with open claw, people making retail payments on an experimental basis. Still not on the enterprise side, as we're talking to many teams. Definitely everyone's looking at it and building infrastructure for when these use cases break through. And so that's part of what we'll be displaying today with Ampersand. But in order for this to really break through, we need a compliance layer, specifically when you're dealing with almost a quadrillion-dollar industry of the traditional financial world. There's a set of rules, policies, and guidelines that need to be in place before anything can happen. Beyond just a credit card transaction, we're dealing with trillions and trillions of dollars being transacted across the globe. And so we're going to be honing in a bit on that as part of the presentation today and the demo that's coming up. But the compliance layer is literally, is this counterparty a non-sanctioned entity? Has this counterparty been involved in terrorist activity? Who is the identity behind this? And when you're dealing with the Gentic systems, you're simply presented with just a wallet address with no background information on that. And we need these infrastructure layers in place to enable and facilitate enterprises to feel comfortable adopting this technology. All of these systems exist in the existing world today. But as we're breaking through into this Gentic world, we really need to have all of these pieces before large enterprises are going to sign off and give the okay to implement these. At the end of the day, there is going to be a human responsible, a chief legal officer, chief policy officer, and that will have to sign off. And that person needs to feel 100% confident that the systems in place will not allow for agents to hallucinate, to go off the rails, to overspend, to break policy. A lot of these compliance issues deal with large fines into the tens, hundreds, even billions of dollars. So these are really important items and things to think about from a large enterprise and financial services perspective. And governance simply hasn't caught up. Again, we're early days. So we are part of building up what needs to be put in place from our view in terms of allowing this to scale. And that's why we're building ampersand. So we're dealing with Gentic checkouts with a financial harness. And I'm going to turn it over to my colleague Pranav, who is going to show us a demo. All right? All right. Thanks. All right. Like we said, for agents to be really useful, we need to give them tools. Right now they're being used for coding. If you want to use them to build better things, you need to give them the right tools. As of right now, as an AI industry, we've given them a lot of MCP servers. But most of these MCP servers are free of cost. So there are two ways to make your agents really useful. One way is you can go to all these websites like Exa, Firecrawl, and so many more, which are getting bigger and bigger with more MCP servers coming to play. And you've got to put in your credit card and make your agent better. Or there is a better way that you can use an aggregator, which has all the important tools to make your agent super powerful. And that's why we built this marketplace. And all you need is a skill file installed in your agent. You can give your cloud code the skill file, and everything else will be done for you. So I'm going to show you an experiment. We'll have this specific terminal with the ampersand skill file, and we'll have cloud code, cloud coworking without the skill file. And you'll see the difference is immense already. The reason is that most important MCP servers are going to be paid, and you'll not put in your credit card to all these MCP servers. Rather, what you'll do is you'll use an agentic commerce tool or a platform or a wallet or a credit card to make that happen. And that's what we enable. So let's just go on the terminal and install the ampersand skill file in one set. This is our website, by the way, where you can go, copy the prompt, and just put it in here. I already have it, so I'm not going to maybe, it's fine. Install the skill file. And here I'm on the cloud cowork, right? It does not have the skill file. Paiden square. This has the skill file. Now, what I'll do is I'll tell them, find the email information of the head of crypto and blockchain at MasterCard, right? And I'll put in the same prompt. I'll put in the same prompt for cloud cowork. Remember, the only difference is this specific terminal of mine has the skill file, which has paid MCP tools, and this one does not. And you'll see that it is not able to specifically find out, hey, we're not able to give you exactly the email, but this is how the email format is for our MasterCard, and you can go ahead and find that out. But when I go to my terminal, it will already be able to give me the email specifically, the Twitter, and where is he based, and everything and beyond. This is just one example of telling you that agents are getting powerful through MCP, but these MCPs are not public services. They're going to be charged, and you will not know that you're interacting with MCPs or installing it or going on their websites and getting the credit card. Rather, you'll just have a skill file, which is an aggregator, which can take care of your payments. In the background, what happened was, if you go over here, you already can see that we did a transaction so that we could enable this search for you because this is a paid endpoint that we had to specifically pay for to get what you specifically needed. So when you go to transactions, you should be able to find that we did a small transaction so that you would specifically get that specific information. Maybe we skip to the next demo for now. As you can see, this pending happened. Let's skip to the next demo. I wanted to show you one more demo. In this, what I've done is, let's say I'm not a good kid, which generally I am, and I want to show you one more demo. I want to buy a Father's Day gift for my dad, and I can also make that happen directly via the terminal. Shopify introduced UCP. Amazon might be coming up with its own thing, but what I'll do is I'll give it a specific command. Using Shopify UCP, buy my father a Father's Day gift, keep the gift less than $10. Let's see how this specific thing rolls. What will happen in this specific terminal case is that it will already be able to locate certain shops online and find out what things can be given for Father's Day gift, list them for me, and be able to buy that directly via ampersand wallet that we have created. Your agent is as powerful as the paid MCP tools that you are connected to it, and if you have given it a payment trail. So currently it's finding that out for us. about this specific thing is that again and again I don't have to put in my name, my address, my phone number, anything. The agent already has the memory where it knows what do I like, what do I not like, and it will just give me specific things that I might want to buy for him. So it already, total $9, what it is, confirming the order, and it already plays the bet, and it already took $11 for that. Order complete, and here is your receipt, and my dad gets his $10 key directly via the agent. So this is just starting of what we call will be the future of agent e-commerce. It's not just buying gifts for your parents, but rather it's much bigger. Here, MCPs are becoming the normal norms of how your agents become super powerful. And you might have already seen that Cloudflare is opening its gateway through X402 and agentic payments. If you go on a website and the agent is crawling that website, then ads are irrelevant. How do people get paid? That's why our agentic payments, where a bot comes through Cloudflare, pays a microtransaction, and gets all the information that's needed. So to make your agent super powerful, you need paid tools. We are in the era where most MCP tools are free, but that's not going to be the case in the future. They're going to get paid, and what your agent would need is good MCPs or an API which has an integration of MCPs and a wallet. And this is what I'm trying to show you. So the specific bet has already been paid. I'm going to try to show you the transaction over here that was done, and it's already settled. And I might have gotten an email where Shopify was able to buy my dad a gift. Let's see if I open it and show it to you because, yeah, it's okay. You guys are friends. See, over here, the order has been placed, and it's already getting tracked all via the agent terminal. I didn't have to even skip. Let's go to the last part, which is all of this will be big if and only if, like Rodrigo says, there is compliance. Merchants will not take payments if they think this order is being placed by North Korean wallet. And for that, you need the compliance layer to be fitting in as well. The seller and the buyer side both need to have that. And I've created this specific simulation for you in which we have a good claw that you can let me just close this right now. So now, let's go to the third one, which is compliant transactions. Right? For the world and Amazon to accept agentic payments, there needs to be compliance involved to make sure that the payment that's been done is not malicious. So we have a good claw that I'll be spinning up over here, and we have a bad claw. The thing about bad claw is this is, again, a simulation, but this is a sanctioned wallet address. That means it has either interacted with North Korean entities or we were able to simulate it in a way that it is flagged. Right? So no amount of transaction should be able to be done because it is out of the swift scope of policies. So if we go over here in this specific thing and I disable screening and save it, my good claw and my bad claw both should be able to do transactions. Just to give you a little bit of periphery, this is the seller that is accepting payments. We have a small service of scraping websites, and it needs to pay 0.1 cent to be able to take that scraping information. Right now, because we don't have regulations, both these transactions are going through, and you can see it's getting authorized over here. Right? Now what I'll do is, using compliance, I will enable this specific feature that we have built with TRM, which scans the wallet and makes sure that all the transactions are compliant. And what it will do is that up till now, everything has been working for good claw and bad claw. But things will stop working for bad claw because, using compliance, we are able to scan that and we are able to give you, see over here, your transactions are getting blocked or rejected. And I can also show you over here that rejected and denied. And the reason behind the denying is because you're in the block-listed wallet addresses. All to tell you that agentic commerce is becoming real. Agents need commerce more than humans need commerce. But it will not be the same. You will not go through payment guardrails of Stripe or any other of these checkout flows. But agents will have their own proprietary forms, either through wallets or through their own credit cards. Agents will need to be superpowered. MCPs will get paid. And more and more paid MCPs will come to reality to make your agent successful. That will only happen and only be successful in UX terms and more if you're able to empower your wallet with paid MCPs and a wallet infrastructure. If you want to know more about it, come to Ampersand. We'll be here outside chatting if you have any questions. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. They're going to be the case in the future. They're going to get paid, and what your agent would need is good MCPs or an API which has an integration of MCPs and a wallet. And this is what I'm trying to show you. So the specific bet has already been paid. I'm going to try to show you the transaction over here that was done, and it's already settled. And I might have gotten an email where Shopify was able to buy my dad a gift. Let's see if I open it and show it to you because, yeah, it's okay. You guys are friends. See, over here, the order has been placed, and it's already getting tracked all via the agent terminal. I didn't have to even skip. Let's go to the last part, which is all of this will be big if and only if, like Rodrigo says, there is compliance. Merchants will not take payments if they think this order is being placed by North Korean wallet. And for that, you need the compliance layer to be fitting in as well. The seller and the buyer side both need to have that. And I've created this specific simulation for you in which we have a good claw that you can – let me just close this right now. So now, let's go to the third one, which is compliant transactions. Right? For the world and Amazon to accept agentic payments, there needs to be compliance involved to make sure that the payment that's been done is not malicious. So we have a good claw that I'll be spinning up over here, and we have a bad claw. The thing about bad claw is this is, again, a simulation, but this is a sanctioned wallet address. That means it has either interacted with North Korean entities or we were able to simulate it in a way that it is flagged. Right? So no amount of transaction should be able to be done because it is out of the swift scope of policies. So if we go over here in this specific thing and I disable screening and save it, my good claw and my bad claw both should be able to do transactions. Just to give you a little bit of periphery, this is the seller that is accepting payments. We have a small service of scraping websites, and it needs to pay 0.1 cent to be able to take that scraping information. Right now, because we don't have regulations, both these transactions are going through, and you can see it's getting authorized over here. Right? Now what I'll do is using compliance, I will enable this specific feature that we have built with TRM, which scans the wallet and makes sure that all the transactions are compliant. And what it will do is that up till now, everything has been working for good claw and bad claw. But things will stop working for bad claw because using compliance, we are able to scan that and we are able to give you, see over here, your transactions are getting blocked or rejected. And I can also show you over here that rejected and denied. And the reason behind the denying is because you're in the block listed wallet addresses. All to tell you that agentic commerce is becoming real. Agents need commerce more than humans need commerce. But it will not be the same. You will not go through payment guardrails of Stripe or any other of these checkout flows. But agents will have their own proprietary forms, either through wallets or through their own credit cards. Agents will need to be super powered. MCPs will get paid. And more and more paid MCPs will come to reality to make your agent successful. That will only happen and only be successful in UX terms and more. If you're able to empower your wallet with paid MCPs and a wallet infrastructure. If you want to know more about it, come to Ampersand. We'll be here outside chatting if you have any questions. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you. Thank you.