Exploring the New Trust Infrastructure of Autonomous Agent Economy
Written by: Decentralised.co
Translated by: AididaoJP, Foresight News
In the article "Internet Pricing," we discussed that when measuring payments is frictionless, machines will automatically make payments. Humans have not fully embraced micropayments because focusing on the measurement process requires effort and mental energy. Machines, however, only see 1s and 0s. Mental capacity or task switching does not affect their execution ability. If breaking down to sub-cent levels makes the process more efficient, they will do so, which is different from humans.
We ended the previous article with a question: What happens when agents mess things up? The correctness of the agent's intent does not matter. The key is that we cannot supervise agents at every step.
This puts us in a dilemma: new technologies have failed to inherit a significant advantage of old infrastructure, such as the ability to reverse payments when errors occur. This article will explore this issue. We will discuss what is needed for agents to achieve autonomy, who is building the infrastructure for this, and why new companies are emerging at the intersection of blockchain payment channels and autonomous agents.
Emerging Standards
Any commercial activity involves three parties: the buyer, the seller, and the intermediary facilitating the transaction. The intermediary can be a platform or marketplace like Amazon, or a card organization network like Visa that processes payments.
Buyer
Consumer applications typically handle funds or transactions and take a cut from them. But what happens when the consumer is an AI acting on our behalf? Currently, several emerging standards are seeking answers.
ChatGPT has 700 million active users, all trying to obtain information or services through AI. While we have not yet directly bought and sold goods through agent interfaces, we commonly use it to "discover" products. Whether buying running shoes or finding hotels in El Calafate, I am comparing prices using AI. It would undoubtedly be much more convenient to purchase directly through the same interface. This is precisely the purpose of OpenAI's collaboration with Stripe to launch the Autonomous Agent Commercial Protocol (ACP).
This is currently the most direct way for agents to handle funds: users have full control throughout the process. After the user places an order, ChatGPT sends the necessary information to the merchant's backend via ACP. The merchant then decides whether to accept or reject the order, processes the payment through the existing payment service provider, and handles shipping and customer service as usual.
You can think of the ACP business model as: you authorize an intern to spend a fixed budget, and you ultimately decide which product/service to purchase and from which merchant to complete the payment.
OpenAI and Stripe have ACP, while Google has introduced the Agent Payment Protocol (AP2). Before delving into AP2, let's take a step back. What Google wants to solve is the "interoperability" issue. Currently, AI agents operate in silos: Gemini does not communicate with Claude, and ChatGPT does not know what is happening in Perplexity.
Ideally, when tasks become complex and require collaboration, we want these agents to communicate in a common language. To this end, Google has developed A2A (Agent-to-Agent Protocol) to enable different agents to communicate and coordinate.
But just being able to converse is not enough. Agents also need to use tools, access APIs, and services. The Model Context Protocol (MCP) allows agents to use tools like Google Calendar, Notion, and Figma.
MCP defines a universal language. As long as they "speak" MCP, agents can use any tool without additional custom code. The protocol was created by Anthropic, but the specifications are open and are being rapidly adopted by various companies. MCP servers essentially act as a translation layer, sitting in front of a company's existing APIs, exposing services in a standardized format to any agent compatible with MCP.
Returning to AP2, it can be simply understood as follows: MCP gives agents the ability to access data, files, and tools; A2A gives them a voice to communicate with each other; and AP2 gives them a wallet to spend securely.
All these protocols place users at the center of control, with agents having limited spending authority. This addresses distribution and process issues, but it still does not solve the question: what happens when agents make mistakes?
Seller
The story does not only happen on the buyer's side. Sellers are also emerging with new standards, focusing on how machines pay for access to APIs, data, and content.
The most discussed currently is the x402 standard, an open protocol developed by Coinbase. It resurrects the HTTP status code 402—"Payment Required," which was defined back in 1997 but never used. x402 revives this status code by combining it with stablecoin payments, allowing micropayments to be settled economically.
x402 turns HTTP requests into paid requests. Whenever payment is required, the server makes a request. Since agents have a preset budget, they will pay the server and obtain data within the same process. This makes "pay-per-request" or "pay-per-call" feasible in machine-to-machine transactions.
With x402, agents can pay precisely for what they need at the moment. For example, paying 2 cents to read a paid article or a fraction of a cent for an API call. Transactions can be settled on-chain within seconds, without establishing long-term relationships.
Cloudflare has borrowed this concept to build a more specific "pay-per-fetch" system. It also uses HTTP 402 at its core, but the key lies in Cloudflare's market dominance, as 20% of global internet traffic passes through its network, giving it significant influence.
"Pay-per-fetch" utilizes Cloudflare's edge network, requiring payment before providing content to AI crawlers. This turns access to content into mandatory measurement. Publishers are facing a traffic drop because people no longer click into websites from search engines but directly read AI-generated summaries. Through this system, publishers can charge AI labs directly each time a crawler accesses their content.
Card organizations are also trying to extend existing payment channels to handle agent transactions. Visa has launched MCP servers and acceptance agent toolkits. Mastercard has a project called "Agent Payments." Both are in early pilot stages, but they are important because Visa and Mastercard already have global distribution networks, issuer relationships, and extensive merchant acceptance networks. The basic idea is to register agents, set spending controls, and allow agents to initiate transactions on existing human credit card payment networks.
Bridging the Trust Gap
All the aforementioned standards assume that payments will proceed smoothly and outcomes will meet expectations. ACP and AP2 involve human participation at the checkout stage, providing a certain level of security. The x402 variant deals with machine-to-machine data access, where risks are generally lower. Card organizations extend their familiar protective mechanisms, but at the cost of slow settlements and high fees.
To achieve large-scale micropayments, speed is the primary goal. Card payment networks take days to settle, and merchants must pay a percentage of the transaction amount as fees. Cryptocurrency channels can settle in seconds at a cost of less than a cent. However, this efficiency comes with irreversibility; once a cryptocurrency payment is made, it cannot be reversed.
Traditional commerce has built an entire infrastructure around "what could go wrong." When credit card shopping goes awry, you have processes to follow: contact the bank, initiate a dispute, the card organization investigates and temporarily holds funds, and ultimately decides on a refund or supports the merchant. In 2025, there were 261 million disputed transactions, totaling $34 billion.
However, agents operating on stablecoin channels have none of these safeguards.
When agents begin to collaborate, the issues become more complex. When hundreds or thousands of multi-agent workflows intertwine, clarifying responsibility can become a nightmare.
Card organizations will not bear this risk, at least not under the current profit model. Visa and Mastercard's agent projects still charge standard interchange fees, and settlements still take days. They could turn to instant stablecoin settlements, but that would mean abandoning the dispute resolution system that forms the basis of their fees.
The dispute resolution mechanisms of traditional finance were not inherent. The first credit card (Diners Club) emerged around 1950, but consumers had to wait 24 years to gain transaction dispute rights. The modern infrastructure we take for granted today was built gradually as issues arose.
The autonomous agent business does not have the luxury of time to waste. API requests now account for 60% of the dynamic HTTP traffic processed by Cloudflare. Bot and automated traffic now make up nearly half of internet traffic. ChatGPT's 700 million users can already check out directly on Etsy through ACP, and Shopify integration is coming soon. The transaction volume already exists, and users have a potential need for agents to handle tasks; it is not far-fetched for agents to be used in commercial activities.
Therefore, we face a choice: should we let traditional financial infrastructure continue its slow settlements, or should we consciously build a trust infrastructure to match the rapid blockchain settlements? The former will limit the potential of agents, while the latter is an opportunity and an inevitable extension of the development of autonomous agent commerce.
So, what specifically should be done?
Unsurprisingly, this involves two parts: pre-transaction and post-transaction.
Pre-Transaction: Should Agent Transactions Be Allowed?
This depends on three points: identifying counterparties, fraud detection, and using reputation scores to determine pricing and access permissions.
In the U.S., Plaid connects nearly half of bank accounts, processing millions of account verifications daily. When you verify your identity on Venmo, you are using Plaid.
Currently, any agent interacting with APIs, scraping web pages, or initiating payments lacks peer identity verification. The server only sees a vague ID (like a wallet address or API key) and does not know who the caller is. Without a universal identity across services, reputation cannot be accumulated, and each interaction starts from "zero trust."
In 2024, American adults lost approximately $47 billion due to identity fraud.
We need a "Know Your Agent" (KYA) layer, similar to how Plaid provides identity infrastructure for fintech. It should issue persistent and revocable credentials that bind agents to the humans or organizations behind them.
Card organizations have spent decades building systems that can identify suspicious patterns from millions of transactions. They understand normal human consumption behavior and can flag anomalies in real-time. If an agent is compromised and makes unauthorized purchases across multiple merchants, there is currently no shared fraud map to detect it.
Visa claims that after investing $11 billion to enhance security from 2019 to 2024, its systems prevented $40 billion in fraudulent attempts. Stripe processes over $1.4 trillion in payments annually and trains its Radar anti-fraud system based on this data. During Black Friday and Cyber Monday in 2024, Radar blocked 20.9 million fraudulent transactions worth $917 million.
Agent transactions currently lack such a fraud detection layer. When an agent makes an x402 payment, there is no shared system to flag abnormal behavior, such as a surge in spending or unusual frequency.
Without persistent identity and reputation, each interaction by an agent starts from scratch. Reputation is deeply embedded in human commerce: the ads you see are based on browsing history, Uber ratings affect driver acceptance, and credit scores follow you to every financial institution. The same should apply to agents.
Post-Transaction: What Happens When Things Go Wrong?
Chargebacks are how card networks handle disputes: after a customer disputes a transaction through their bank, funds are withdrawn from the merchant. However, this is often abused. In 2023, chargebacks caused merchants approximately $117.47 billion in losses. For every $1 lost in refunds, merchants typically incur an additional cost of $3.75 to $4.61 (including fees, product losses, and administrative expenses).
Source: Coinbase's x402 paper
Merchants win only 8.1% of disputes they actively contest. 84% of customers believe that initiating a chargeback directly with the bank is easier than seeking a refund from the merchant.
Stablecoin transactions initiated by agents settle in seconds and currently cannot be reversed. Cloudflare has proposed a delayed settlement extension for x402, allowing a "waiting period" before funds are finally transferred.
Developers are already building prototypes of this infrastructure. At the ETHGlobal Buenos Aires hackathon, a team created Private-Escrow x402. Their escrow solution involves buyers prepaying funds into a smart contract and signing a "payment intent" off-chain at the time of payment. A coordinator batches hundreds of such signatures into a single settlement transaction, reducing Gas fees by 28 times.
But this is just a foundational component; it still needs to be productized.
Who Will Build All This?
This reminds me of the era when telecom operators dominated the industry. They had billing relationships with every mobile user but missed out on the value generated by smartphones. App distribution and mobile advertising created hundreds of billions in revenue that could have been captured by operators.
Card organizations now face a similar situation. The trust infrastructure that Visa and Mastercard have built over decades is precisely what the autonomous agent economy lacks. However, their business model relies entirely on interchange fees, which exist on the premise that they control payment channels. They spend heavily to maintain this infrastructure, funded by a few percentage points of transaction volume. Providing consumer protection for stablecoin transactions would mean subsidizing competitors' payment channels with their own revenue.
If card organizations do not act, the next candidates are AI labs like OpenAI, Google, and Anthropic. They all want their agents to be widely used. However, operating centralized identity registration means they would have to take responsibility when agents behave improperly. They do not want to be the adjudicators when you "book the wrong hotel."
They would prefer third parties to build identity and recourse infrastructure for them to connect directly, just as they do today with payment or search engines.
Cloudflare is in a unique position. They have processed massive amounts of internet traffic and run crawler detection, with their "AI audit" tool allowing publishers to track crawler access. Transitioning from "identifying bots" to "validating agent identity and reputation" is not a significant technical leap.
However, Cloudflare has always prided itself on being a neutral infrastructure. Once it starts issuing trust scores or adjudicating disputes, it becomes more like a regulatory body—this is a different business and implies different responsibilities.
Three Entry Points for Startups
You cannot outdo OpenAI in model quality, nor can you surpass Cloudflare in traffic. You need to find parts of the tech stack where their business models (at least for now) do not allow them to touch, yet still hold value. I believe there are three entry points: identity, recourse, and attribution.
Agent identity is the most direct. The registration model has been validated. While Plaid is a classic case, it is quite fitting: they provide identity verification for bank accounts. Startups can do the same for agents: issuing credentials, accumulating reputation, and allowing merchants to verify reputation scores before accepting payments. Their moat comes from network effects: once enough merchants verify through your registration form, agents will have to maintain a good reputation record.
Recourse mechanisms are more challenging because they require taking on risk. You can think of it as insurance: charging a small fee for each transaction and covering losses when things go wrong. Scale is key. Card interchange rates are 1.5% to 3%, which includes dispute handling costs. The costs of stablecoin channels are far lower, so a recourse layer could easily provide comparable protection at a 0.5% rate and still have room for profit.
Attribution mechanisms are the most forward-looking but will inevitably emerge. As agents begin to influence purchasing decisions, brands will pay to influence recommended content. Auction mechanisms can be designed. However, it has a "cold start" problem, requiring brands, agents, and merchants to participate in the market for it to function, which the first two entry points do not face.
The importance of these three entry points changes with the development stage of the agent economy:
Identity becomes critical when agents no longer require manual approval for each transaction.
Recourse is essential when agents start handling real funds.
Attribution will only kick in when the transaction volume between agents is sufficient to support an advertising market.
This leads to a practical development trajectory:
Source—Chart generated using Claude
Startups Will Build Parts of the Agent Economy Infrastructure
The development of agents can be divided into three stages:
As an interactive interface
Executing under human supervision
Autonomous trading with each other
We are currently in the first stage. ChatGPT's Etsy checkout integration is a good example: we browse products in a chat interface (though not exclusively), agents recommend options, but the final decision is made by humans. Trust is entirely borrowed from existing infrastructure.
This stage belongs to existing giants, as it is a distribution game for user entry points. Value accumulation is in the hands of players who have the purchasing decision interface.
The second stage is marked by agents gaining more autonomy. Agents no longer just suggest itineraries but directly book flights, rent cars, and reserve hotels. We provide goals or constraints, agents execute, and we accept the results.
At this point, the trust layer becomes indispensable. Without a recourse mechanism, users will not authorize agents; without identity verification, merchants will not accept agent payments.
This is precisely where the opportunity lies for startups. Existing giants may lack sufficient motivation to build trust infrastructure for stablecoin channels, as they already have significant growth potential in the current stage (still led by themselves). OpenAI's revenue reached $13 billion this year. In contrast, Tether's profits in just the first ten months of 2025 are expected to reach $10 billion, with even higher profits for the entire year.
The layers of identity, recourse, and attribution will be built by new companies dedicated to solving the specific issues of agent capabilities and user authorization boundaries.
The third stage is autonomous agent commerce. Your agent will not need to consult for everyday decisions; it can negotiate with other agents, bid for computing resources, participate in advertising auctions, and continuously settle thousands of small transactions. Stablecoins will become the default settlement layer due to their capacity, speed, and granularity required for machine-to-machine transactions.
The competitive focus in this stage will no longer be on the best model or the fastest public chain, but on who builds the most trusted infrastructure: the agents' "passports," the "courts" for adjudicating disputes, and the "credit systems" that allow over-balance transactions. These institutions providing software services will determine which agents can participate in the economy under what conditions.
Conclusion
We have paved the way for agents to "spend money," but we have yet to establish mechanisms to verify "whether they should spend." HTTP 402 has been dormant for thirty years, only awakening now due to the feasibility of micropayments. The technical issues have been resolved. However, the trust infrastructure supporting human commerce, such as identity verification, fraud detection, and dispute resolution, still lacks corresponding agent versions. We have solved the easier parts. It will take time to enable agents to conduct business with confidence.
免责声明：本文章仅代表作者个人观点，不代表本平台的立场和观点。本文章仅供信息分享，不构成对任何人的任何投资建议。用户与作者之间的任何争议，与本平台无关。如网页中刊载的文章或图片涉及侵权，请提供相关的权利证明和身份证明发送邮件到support@aicoin.com，本平台相关工作人员将会进行核查。
