With the AI real-time competition of Nof1 and the x402 protocol launched by Coinbase becoming hot topics, the application scenarios of AIAgent in the financial and payment fields are continuously expanding. As a representative of AI payment protocols, how does the x402 protocol differ from traditional payment protocols? What are its payment scenarios? And as AI payment improves, what foreseeable application scenarios can AI Agent have in the future? To address these questions, we invited Yang Haipo, founder and CEO of ViaBTC, to engage in an in-depth discussion about the feasibility of the x402 protocol and the future imagination of AI collaborative networks.
Q: The x402 protocol has recently become a hot topic in the industry. What is your perspective on using Token payments to solve AI payment issues with the x402 protocol?
Yang Haipo: Actually, setting aside the concept of "AI payment," from an engineer's perspective, x402 is a relatively simple protocol. Its core is not to invent a new payment method but to package on-chain payments as a standard Web service, introducing a Facilitator to solve the trust and execution issues of on-chain payments.
Many people compare x402 with traditional payments, but they serve different targets. Traditional payment experiences like Alipay and Visa are indeed excellent, but they are designed for humans, not for AI. For AI Agents, there are two obvious problems with the traditional payment system: the first is the entry barrier; it's difficult to open a bank account and complete KYC with a script, but generating a wallet address for on-chain payments only requires one line of code. The second is friction costs; AI interactions are high-frequency and fragmented. For example, if an Agent calls a data interface and needs to pay $0.0001, the bank fees through Visa might exceed that amount.
So, x402 actually utilizes the programmability of Tokens, combined with the role of the Facilitator, to solve the problem of "automated micropayments." In this scenario, the Facilitator acts like "Alipay" in the machine world, digesting complex on-chain confirmations, allowing AI to complete high-frequency transactions at millisecond speeds.
In traditional on-chain payments, interactions are slow and complex. The idea of x402 is to have the Facilitator act as the "execution agent" for on-chain transactions, responsible for verifying signatures, covering Gas fees, submitting transactions, and handling on-chain details, while the payer only needs to submit the signature to the Facilitator without directly completing on-chain operations. For both buyers and sellers, this greatly simplifies the payment steps, as the Facilitator resolves trust and settlement issues for them.
Q: What is your view on the development prospects of x402? What problems or limitations might it face in the actual implementation process?
Yang Haipo: If we talk about prospects, I believe the future value of x402 mainly lies in the Agent-to-Agent economic network, rather than the payment experience for end users. For ordinary users, payments should be completely seamless. In the future, you won't see AI Agents prompting you to "scan to pay." When you give the command "help me analyze market trends every morning at nine," the Agent will automatically call news or social media data from multiple service providers in the background. Facing the potential costs of high-frequency calls, the Agent can autonomously pay through the x402 protocol and obtain services without any human intervention. This model will drive API calls from the traditional "membership subscription" model to a true "pay-per-use" model, as x402 is inherently suitable for high-frequency, fragmented collaboration scenarios between machines.
Additionally, there is a security advantage that is easily overlooked. Today, if you want AI to help you buy something, you basically wouldn't dare to give it your Visa card number because credit cards essentially carry unlimited liability. If the Agent is attacked or has a hallucination, it could indeed "max out your card." However, a Token wallet can set an authorization limit for AI, such as a "pocket money account" of 100 USDC. This way, even if the Agent encounters issues, the losses are controllable.
However, because x402 is designed to be so simple, its shortcomings are also very apparent. First, the x402 protocol is highly dependent on Facilitators like Coinbase. It simplifies development but also introduces single-point risks. If a Facilitator's server goes down, or if it acts maliciously or censors your transactions, the entire payment link is broken, which is a typical centralization risk. On the other hand, the simplicity of x402 also leads to certain functional deficiencies, such as "refunds." Currently, the x402 protocol does not have a built-in refund mechanism. Real-world commercial payments need to handle a lot of disputes, such as incomplete services or damaged goods, and the irreversible nature of x402 makes it difficult to implement these processes.
In this context, the industry is also exploring broader and more universal Agent payment protocols, such as Google's AP2, which is attempting to establish a unified standard that can accommodate traditional Visa/Mastercard, as well as cryptocurrencies, while also considering complex commercial processes like refunds. In the long run, large protocols like AP2 are certainly the direction we hope to see, but due to their complex design and the many stakeholders involved, they are still some distance from actual implementation. x402, on the other hand, excels in its simplicity; it doesn't need to wait for banks to upgrade their systems—if there are wallets and code, it can be used today.
Q: Returning to the present, from your personal observation, in which fields are AI Agents currently making real-world impacts and generating actual value?
Yang Haipo: To be frank, the biggest beneficiaries of AI Agents right now are actually the developers themselves. The AI pair programming model has become a daily routine for many engineers, and Agents like Cursor have been widely adopted. For large, complex projects, developers certainly won't hand over full responsibility to AI; that is unrealistic at this stage. However, for some tedious yet time-consuming tasks, such as Code Review, unit testing, or even generating parts of algorithm logic, AI Agents can already take on a significant portion of the workload, greatly saving developers' time.
Another noteworthy scenario is the assistance to non-technical personnel. For example, the recently popular "Vibe Coding" has gained traction because it opens up the imagination of non-technical individuals. Previously, if you lacked programming skills, even if you had good ideas, you couldn't realize them. But now, you can communicate your ideas to the Agent in natural language and have it write the code for you. Of course, we must be realistic and say that Vibe Coding is not a "one-click generation" universal tool; the output from the Agent often requires repeated debugging. Additionally, while it can achieve rapid prototyping, after multiple iterations, the code can become bloated and chaotic, posing significant challenges for subsequent maintenance and upgrades. Nevertheless, even with a current success rate of only 30-40%, the ability it provides to non-technical personnel to go from 0 to 1 is still highly valuable.
There are also some seemingly minor but very common needs in actual work. For instance, developers occasionally need an icon, a button style, or a simple interface sketch, and previously they could only ask design colleagues for help. Now, they can have the Agent quickly generate an icon or a draft UI, and even just a "usable draft" can save a lot of back-and-forth communication time.
Although AI often doesn't perform perfectly at this stage, it is already good enough for small teams or independent developers looking to create a demo or MVP.
Q: From your observation, where do you see the future potential of AI Agents? Is it possible that similar new attempts will emerge in the crypto industry?
Yang Haipo: Looking at a longer time frame, I believe the potential of AI Agents certainly extends beyond the current level of development assistance. There are many future possibilities, such as more autonomous collaboration and procurement tasks that could be realized.
Currently, there are already some interesting explorations in the industry. For example, the AI real-time competition initiated by Nof1 essentially allows Agents under different models to test their strategy capabilities in real market environments. In this case, AI is no longer just providing information to humans but is forming its own action loop.
Moreover, more and more exchanges are beginning to support MCP (Model Context Protocol), such as CoinEx, which is part of our ecosystem and has already released the corresponding MCP service on GitHub. With such MCP services, AI Agents can directly access real-time market data, candlestick data, news feeds, etc., and perform in-depth analysis in conjunction with models. Theoretically, it can not only automatically generate strategies based on user preferences and risk parameters but also, if deployed locally, it can have the capability to place orders automatically. In such scenarios, AI Agents will truly possess capabilities for automated trading and intelligent market making. For example, they can dynamically adjust order prices and quantities by real-time acquiring market depth, volatility, and trading volume data, thereby improving market efficiency and liquidity. The emergence of such capabilities marks the transition of Agents from merely "helping you look up information" to "helping you make decisions and execute them."
In this model, we can also see the application of x402. For instance, if you ask your Agent A to write a deep analysis report on Bitcoin, and it lacks relevant data, it will automatically call other Agents to complete the entire process: requesting on-chain position and transaction data from Agent B, which is responsible for on-chain data monitoring, and automatically paying; requesting sentiment summaries from Agent C, which is responsible for news aggregation, and completing another small payment. What you see as just "receiving a report" actually involves multiple microtransactions between Agents.
From the existing examples, we can see that Nof1 has proven that AI can make decisions, MCP has solved how AI can acquire data and execute, and x402 enables AI to collaborate economically with other Agents. Therefore, I believe the potential of AI Agents will manifest in two directions: one is stronger autonomous decision-making; the other is more natural economic collaboration. When Agents can automatically seek resources, purchase services, call tools, and complete the entire task chain, what we see is no longer the capability of a single model but a digital economic system composed of multiple Agents. These developments may just be starting today, but I believe the future of AI Agents is full of possibilities.
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。
