Author: Kevin, Movemaker Researcher; Source: X,@MovemakerCN

Google’s recently released AP2 agreement has set the underlying rules for payment and transactions for the upcoming Agent economy.Its core task is to complete the last section of “AI analysis” (based on MPC) and “Agent execution” (based on A2A), “value settlement”, trying to normalize the payment process standards of AI Agent, and ultimately promote AI Agents that can independently complete business activities on behalf of users.

AP2’s primary application scenario is aimed at the core of traditional Internet such as e-commerce and subscription services.This means that its original intention is to solve the problem of how to safely delegate payment authorization from humans to AI under the existing business framework.This will directly affect the direction of e-commerce operations, and users will be freed from clicking on “buy” and “subscribe” operations and transform to “intentional delegation” and “rule setting” of AI Agent.

It is based on this logic that we need to evaluate the impact of AP2 on the crypto industry more carefully.It is not born for Web3 native scenarios, but incorporates crypto payments (particularly stablecoins) into its grand framework as an optional, integrated payment option.Therefore, rather than saying that AP2 will trigger an explosion of AI, it is better to say that it provides an interface for crypto assets to enter mainstream AI applications.

Understand the pre-protocols of AP2: MCP and A2A

The existence of AP2 is based on the development path of the two pre-protocols, MCP and A2A.Understanding the nature and limitations of these two protocols is the key to understanding the true intentions of AP2.MCP solves the problem of connecting AI with tools, while A2A tries to solve the problem of collaboration between AI and AI.

What is MCP?Let AI truly touch the real world

The essence of large language models (such as Gemini, GPT) is a “super brain” isolated from the real world.It has a huge amount of static knowledge, but it cannot obtain real-time information and cannot directly operate any external software.Users issue instructions like “Please help me book a ticket to Osaka tomorrow”, which cannot be completed because it does not have the ability to interact with the ticket booking website.

MCP(Model Context Protocol) protocol is proposed to solve the problem of how AI communicates with the external environment.MCP defines a standardized communication format that allows AI to make requests to external tools (such as databases, APIs) and receive returns.It can be understood as installing a telephone for this “secret room brain”.

The MCP (Model Context Protocol) protocol is the “standard intercom system” created to break this secret room.

It is driven by industry giants such as Anthropic and aims to address the problem of how AI interacts with external data and tools such as databases, software, blockchains.Through this “walking radio”, the AI ​​model can make requests to the outside world in a standard language, while the MCP system is responsible for accurately passing these requests to the corresponding external tools (such as the weather API, booking website API), and after the tool is completed, the results are transmitted back to the AI ​​in a standard format.

However,Analyzing the current MCP implementation, we will find that this is more like an “inside phone” that can only call a few preset numbers.

At present, the ability of AI to call external tools is highly dependent on the vertical integration of the platform.For example, Gemini calls Google Search by default and uniquely when it needs to search; its code execution also runs in a sandbox environment built by the platform.This model has a direct impact on users and developers:

• For users: The AI’s capability ceiling is locked by the platform.Users cannot choose what they think is better (for example, using Perplexity for AI searches, or calling specific financial data APIs).The practicality of AI is therefore greatly limited, and the user’s experience is tied to the platform’s ecosystem.

• For developers: This creates a de facto market barrier.Even if developers create AI applications that are far superior to the platform’s own tools in a certain field, they cannot be fairly discovered and called by mainstream AI Agents.Innovation is suppressed, and market competition is limited to a closed, platform-led cooperation system.

A2A: The concept of “inside phone” to “public yellow pages”

If MCP is the “hands and feet” of AI to operate tools, then the A2A (Agent-to-Agent) protocol allows independent AI Agents to learn how to “communicate and collaborate”.It is built on MCP and defines a common set of rules toIndependent AI Agents developed by different companies and developers can discover and understand each other’s abilities and collaborate on complex tasks that a single Agent cannot do..

The significance of A2A is to upgrade the MCP’s task processing mode from “single-soldier combat” to “project subcontracting”.For example, when users propose the complex need to “plan a beach party in Hawaii next month”, under a closed system with MCP alone, individual agents can only clumsyly call the limited tools allowed by the platform one by one.

But in the A2A’s conception, individual agents can play the role of “contractors”.It disassembles the tasks and then publishes requirements in the A2A network, looking for professional “travel agents,” “catering agents,” and “event planning agents” to complete them together.This shift in pattern will have profound implications:

• For users: The AI ​​Agent’s capabilities will achieve an exponential leap.Users can delegate extremely complex, cross-domain tasks with confidence, because AI is no longer a simple operator of a tool, but a coordinator of resources.

• For the industry: The goal of A2A is to create an open, interoperable Agent services market.It is committed to solving three core issues:How to discover(Create a mechanism similar to the Agent app store),How to understand(Standardized description language that defines Agent capabilities) andHow to collaborate(The process of standardizing task allocation and progress synchronization).Once this system matures, any Agent developer who meets the standards can register their services and be fairly selected by the market.

AP2’s core mechanism: establishing a trust and authorization framework for Agent transactions

The MCP and A2A protocols solve the problem of “what can be done” and “how to collaborate”, but this immediately raises a more difficult business challenge: Who will be responsible for the legitimacy and consequences of these actions when Agent begins to perform operations involving funds independently?

The traditional online payment system has a risk control logic based on a basic premise: there is a human user performing active and real-time operations in front of the screen.Any non-human-initiated, automated payment request will be marked as high-risk by the existing financial system.This is the most fundamental obstacle to the formation of the Agent economy.Specifically, an AI-initiated transaction request brings three unanswered questions to merchants and payment networks:

1. Certificate of authorization: How to confirm that the user has truly authorized the AI ​​to conduct this transaction?

2. Intent to be true:Does the specific requests of AI (such as purchase quantity, price) accurately reflect the user’s true intentions, especially when the user does not have real-time monitoring?

3. Responsibility: If the transaction is wrong, who should bear the losses caused?

Google’s AP2 protocol is designed to systematically answer these three questions.AP2 needs to establish an open and standardized set of users, AI Agents and merchants.Authorization and Trust Communication Protocol.

The core technology of AP2 is a set of dual authorization mechanism:

1. Real-time authorization (Cart Mandate)

This mechanism is no stranger to credit card users. When you use credit card to pay on the website, your phone will pop out of the authorization window.This is what AP2 does.

After completing preliminary work such as product research and price comparison, Agent will generate a “shopping cart” containing all transaction details and present it to the user.The execution of the transaction must wait for the user to undergo final review and signature confirmation.

Impact analysis: Real-time authorization is the initial form of Agent payment. It lowers the user’s usage threshold, but it does not fully exert the autonomy of Agent.Its main value lies in optimizing the decision-making process before the transaction, but the final “first kick” still requires human intervention.

2. Intent Mandate

This is a transformative part of AP2.Users can pre-set a complex set of delegated instructions.For example: “Only when the price of Tesla Model Y is reduced by more than 20,000 yuan, the purchase process will be automatically executed and the funds of Account A are used. If Account A is insufficient, Account B will be used.”

In order to make this “foreign delegate” trustworthy and secure, the implementation logic of AP2 is based onVerifiable Credentials (VCs):

• The user’s complex delegation conditions are compiled into a cryptographically signed, tamper-proof digital “transaction contract” (i.e., VC).

• At every key step in performing a task, the AI ​​Agent needs to present this “contract” to prove the legitimacy of its behavior.

• All execution links will also generate corresponding VC records, ensuring the auditability of the entire process.

For merchants and clearing networks (such as Visa): What they received was a “transaction contract” with user encryption signature and clarified the scope and conditions of authorization.By verifying this contract, they can gain enough confidence to confirm the legitimacy of the transaction, thereby significantly reducing the risk of fraud and compliance costs.

For users: This makes the true “7×24-hour independent economic agent” possible, liberating users from high-frequency and repetitive decisions.

In short, the real innovation of AP2 is not to transform clearing networks such as Visa or stablecoins, but to addAdded a trust semantic layer about “who is spending money, why is spending money, and can be traced beyond authorization”.

It is committed to uniformly solve the problem of confirming AI payment intentions in different clearing systems such as fiat currency and encryption, so as to eliminate all participants’ fundamental concerns about the uncontrollable and unverifiable behavior of the Agent.

x402 extension: Service call flow that embeds crypto payment natively into Agent

The core technology of AP2 is to make AI Agent compatible with the existing, human-centric financial authorization system, and the “A2A x402” extension is designed for future native, on-chain A2A payments.x402 expansion is jointly promoted by Google, Coinbase, and the Ethereum Foundation.

The technology selection of x402 comes from an Internet standard that has been idle for a long time – the HTTP 402 status code, which is defined as “Payment Required”.In the past, due to the lack of standardized machine payment methods, this status code has almost never been practically applied.

However, the value of this protocol is highlighted when the main caller of the API changes from human developers to high-frequency, automated AI Agents.

The traditional subscription system or prepaid model seems too bulky and inefficient for the scenario where Agent calls massive services on demand, dynamic, and cross-platform.The core idea of ​​x402 is to solve this problem, which natively couples API calls with payments.

Its workflow is designed to be extremely simple and automated:

1. AI Agent calls a paid service API.

2. The service’s server directly returns a 402 Payment Required response containing the information required to complete the payment (such as the collection address, amount, token type).

3. The Agent built-in wallet or payment module parses this “payment bill” and automatically completes payments on the chain (such as using USDC).

4. Agent carries on-chain payment vouchers and re-initiates API request.

5. After the server verifies payment, it immediately returns the data or service results required by the Agent.

This process seems simple, but it can take advantage of the real-time settlement and highly programmable characteristics of stablecoins, which have an important impact on the business model of AI Agent services.

Impact on AI Agent:

• Realize the true “pay-as-you-can-use”: Agent no longer needs to go through the traditional manual process of “Register an account -> Bind a credit card -> Select a subscription package -> Wait for the service to be activated”.It can dynamically discover any service that complies with the x402 standard in the network and complete payment and use in milliseconds, which provides the basis for Agent’s autonomy and exploration capabilities.

• Extreme cost efficiency: Agent’s payment frequency is much higher than that of humans, and its parallelized task processing capabilities enable it to perform extremely fine-grained micro-payments and stream payments.x402 enables Agent to pay accurately based on the number of API calls, the amount of data returned (Token), and the time calculated, perfectly linking resource consumption with cost, avoiding resource waste in the traditional subscription model.

Impact on AI service providers:

• The protocol layer of the payment system is sinking:x402 directly builds the capability of “access is quotation, payment is service” at the communication protocol layer.Developers no longer need to build complex billing, account and subscription management systems themselves. They can focus on the development of core services, price any API, a data shard, or even a page component in fine-grained pricing and launch it to the market.

• Optimize cash flow and global settlement: With the help of stablecoins, service providers can obtain instant and low-cost global settlement, completely getting rid of the long settlement cycle and high handling fees of traditional cross-border payments.This provides unprecedented convenience for ultra-high frequency and extremely small transaction reconciliation, greatly reducing the threshold for individual developers and small teams to commercialize AI services.

Summary and Outlook: Two-track parallelism, ecology begins to emerge

The value of the AP2 protocol is not the seemingly simple technical details, but the last piece of the puzzle of the commercial closed loop for the technological evolution route of MCP (connection) + A2A (collaboration) –Value settlement.

By empowering Agent to reliable trading capabilities, it laid the first standardized track for the entire industry to transition from the era of “AI tools” to the era of true “Agent economy”.

This layout of Google clearly shows its intention: on the one hand, by being compatible with traditional business, it accelerates the implementation of Agent in mainstream scenarios such as e-commerce and subscriptions, and formulates industry standards,letE-commerce platforms quickly “adapt to” AI intervention;On the other hand, by embracing the encryption native solution, the on-chain Agent-to-Agent payment standards have also been formulated.

Opportunities and Challenges: Crossroads of Encrypted Native Agents

The completion of AP2’s payment capabilities will undoubtedly unlock a large number of application scenarios, such as 7×24-hour independent financial management, enterprise automated procurement and renewal, etc., which will promote the penetration of the Agent-to-Agent economic form.

However, AP2 is a top-down authorization framework driven by tech giants with the primary goal of solving their core business ecosystem problems.Crypto payment plays a role that is “compatible” and “supported” in it.

This has a positive stimulating effect on the encryption native Agent protocol.At present, Agent products that truly focus on Web3 core values ​​such as decentralization, censorship resistance, and privacy protection have not yet achieved key technological breakthroughs, and there is even a lack of clear PMF.

Who can create products with native encryption temperament and innovative design at a time when industry standards are gradually becoming clear is the watershed in the next competition of the crypto AI protocol.

Although the road ahead is unclear, some protocols in the crypto industry have begun to actively lay out around AP2 and the possibilities it brings.From the first partners, we can see a preliminary ecosystem being formed.

Standards and Infrastructure: Coinbase, as the initial proposer of the x402 payment standard, is actively promoting its implementation; the Ethereum Foundation is working hard to formulate underlying protocols such as ERC-8004 (Trust Agent Standard).

Entrance and wallet:MetaMask is committed to creating a self-hosted AI agent wallet and simplifying the user’s introductory process, striving to be a secure entrance to the Agent economy.

Public chain and interoperability:Mesh focuses on optimizing payment routing to ensure the success rate and efficiency of Agent payments.

Payment & Applications: Crossmint, BVNK and other companies provide Agent with multi-channel encryption and fiat currency payment capabilities; while platforms like Questflow already have built-in x402 micro payment system, allowing Agent to get paid according to task results.

Ultimately, the release of AP2 marks the start of the standardization process of the Agent economy.It brings unprecedented room for imagination to the industry, but it also further concentrates the power to control these new economic rules.

For the crypto industry, this is both an excellent opportunity to integrate into the mainstream and accelerate adoption, and it is also a severe test of the dominance of the Agent economic form in the future.How to utilize this infrastructure and embark on a differentiated path with the core value of Web3 will be a track worth paying attention to in the next year.