ViaBTC CEO Haipo Yang: From Nof1 to x402 — A Look at AI Agent Applications and What’s Next

With Nof1’s live AI trading competition and Coinbase’s newly launched x402 protocol becoming major industry talking points, AI Agents are rapidly expanding their use cases across finance and payments.

As a representative protocol for AI payments, how does x402 differ from traditional payment systems? What scenarios does it serve? And as AI payments mature, what other foreseeable applications might AI Agents unlock? This Guest Expert piece summarizes perspectives shared by Haipo Yang, Founder and CEO of ViaBTC, on the feasibility of x402 and the future potential of AI collaboration networks.

Q: x402 has recently become a hot topic in the industry. What is the view on using token payments—like x402—to solve payment problems for AI?

Haipo Yang: From an engineering standpoint, x402 is a relatively simple protocol. Its core value is not inventing a new payment method, but packaging on-chain payments as a standardized web service—and introducing a Facilitator to address trust and execution challenges in on-chain payments.

Many comparisons are made between x402 and traditional payment systems, but these systems serve different “users.” Alipay and Visa offer excellent payment experiences, but they are designed for humans, not for AI Agents. For AI Agents, traditional payment systems currently create two obvious obstacles:

1. 1) High entry barriers: It is difficult for scripts to open bank accounts and complete KYC, while generating a wallet address capable of paying on-chain can be done with a single line of code.
2. 2) High friction costs: AI interactions are high-frequency and fragmented. An Agent might call a data API once and pay $0.0001. Routing that through card networks can introduce fees that exceed the payment itself.

In practice, x402 leverages token programmability—together with the intermediary role of the Facilitator—to enable automated micropayments. In this context, the Facilitator functions like “Alipay for the machine world,” absorbing on-chain confirmation complexity so Agents can complete high-frequency transactions in milliseconds.

In conventional on-chain payments, interactions can be slow and complex. x402’s approach allows a Facilitator to operate as an execution layer for on-chain transactions: verifying signatures, fronting gas, submitting transactions, and handling on-chain details. The payer submits a signature to the Facilitator rather than directly performing on-chain operations. For both buyers and sellers, this reduces integration complexity by centralizing trust and settlement in the Facilitator.

Q: What is the outlook for x402, and what limitations might it face in real-world adoption?

Haipo Yang: x402’s long-term value primarily lies in an Agent-to-Agent economic network rather than consumer-facing payment experiences. For end users, payments should become invisible. In the future, an AI Agent is unlikely to ask a user to “scan to pay.” Instead, a user might set an instruction such as “Analyze the market every morning at 9 a.m.” The Agent could then call multiple service providers in the background for news or social data. Fees generated by high-frequency API calls can be settled automatically through x402, enabling service consumption end-to-end with minimal human intervention.

This model can shift API monetization from subscription memberships to truly pay-as-you-go usage, because x402 naturally fits machine-to-machine collaboration that is high-frequency and highly fragmented.

There is also an often-overlooked security advantage. Allowing an Agent to transact using a credit card number creates effectively unlimited liability. If an Agent is compromised or behaves incorrectly, it could generate uncontrolled spending. With a token wallet, spending limits can be enforced—for example, a capped “pocket money” balance of 100 USDC—keeping potential losses controllable.

However, x402’s simplicity also makes its limitations clear. The protocol relies heavily on Facilitators such as Coinbase. This simplifies development but introduces a centralization risk and a potential single point of failure. If a Facilitator goes offline, behaves maliciously, or censors transactions, the payment flow can break.

In addition, because x402 is designed to be simple, it does not cover certain real-world commerce requirements—such as refunds—within the protocol itself. Disputes around unfinished services or defective goods often require reversals, and irreversibility can make such flows harder to implement.

In parallel, broader Agent payment protocols are being explored, including Google’s AP2, with goals such as accommodating card networks, supporting cryptocurrencies, and handling complex flows like refunds. In the long run, more comprehensive standards may be desirable—but multi-stakeholder complexity can slow deployment. x402’s advantage is immediate usability: a wallet plus code is sufficient to start.

Q: In practice today, where are AI Agents delivering real value?

Haipo Yang: At present, the biggest beneficiaries of AI Agents remain developers. AI pair programming has become routine for many engineers, and tools such as Cursor have seen broad adoption. For large, architecturally complex projects, full responsibility is typically not delegated to Agents at this stage. But for tedious, time-consuming tasks—such as code review, unit testing, and parts of algorithmic logic generation—Agents can meaningfully reduce workload and save time.

Another notable area is enabling non-technical users. “Vibe coding” has attracted attention because it allows people without programming backgrounds to translate ideas into code through natural language. That said, Agent output often requires repeated debugging. Rapid prototyping becomes possible, but after many iterations codebases can become bloated and harder to maintain. Even so, a partial success rate can still be valuable because it enables a 0-to-1 leap for non-technical creators.

Agents are also increasingly useful for small, common workplace needs. For example, generating an icon, a button style, or a simple UI sketch previously required designer support. Agents can now produce quick drafts, reducing back-and-forth and accelerating iteration.

Despite current limitations, these capabilities are already sufficient for small teams and independent developers building demos or MVPs.

Q: Looking ahead, where is the biggest opportunity for AI Agents—and could crypto see similar new experiments?

Haipo Yang: Over a longer cycle, the opportunity for AI Agents is unlikely to remain confined to developer assistance. Future possibilities include more autonomous collaboration and autonomous procurement.

Industry experiments are emerging. For example, Nof1’s live AI trading competition effectively allows Agents built on different models to test strategy capabilities in real market environments. In this setting, Agents move beyond providing information to humans and begin forming closed loops of perception and action.

More exchanges are also starting to support MCP (Model Context Protocol). CoinEx, within the ViaBTC ecosystem, has published an MCP service on GitHub. With MCP services, an Agent can directly access an exchange’s real-time quotes, candlestick (K-line) data, and news feeds, then combine that data with model reasoning for deeper analysis. In principle, an Agent can generate strategies based on a user’s risk preferences and—when deployed locally—can also place orders automatically.

This trajectory enables automated trading and more intelligent market making. By observing real-time market depth, volatility, and trading volume, an Agent can dynamically adjust order prices and sizes, improving market efficiency and liquidity. These developments indicate a shift from “helping with research” to “supporting decisions and execution.”

Within this model, x402 can provide the economic rail for Agent collaboration. For example, an Agent tasked with producing an in-depth Bitcoin research report may lack certain data inputs. It can automatically call other Agents for on-chain position and transaction datasets, or for sentiment summaries aggregated from news, completing micropayments for each service behind the scenes. The end user receives a single report, while multiple Agent-to-Agent microtransactions occur in the background.

Taken together, Nof1 highlights decision-making in live environments, MCP supports data access and execution, and x402 enables economic collaboration among Agents. As Agents become capable of finding resources, purchasing services, invoking tools, and completing full task chains, the result increasingly resembles a digital economic system composed of many cooperating Agents.