在《互聯網定價》一文中，我們曾論述過：當計量支付毫無摩擦時，機器將會自動支付。人類未能完全接納微支付，因為關注計量過程需要耗費精力和心智。但機器不同，它們眼中只有 1 和 0。心智容量或任務切換不會影響其執行能力。如果細分到次美分級別能讓流程更高效，它們就會這麼做，這與人類不同。
上篇文章我們以一個問題結尾：當代理把事情搞砸了怎麼辦？代理的意圖是否正確並不重要。關鍵在於，我們不可能步步監督代理。
這讓我們陷入了一個困境：新技術未能繼承舊有基礎設施的一大優點，例如在出錯時撤銷支付的能力。本文要探討的正是這個問題。我們將討論代理實現自主性需要什麼，誰在為此構建基礎，以及為何會有新創公司出現在區塊鏈支付通道與自主代理的交匯處。
任何商業活動都涉及三方：買方、賣方，以及促成交易的中間方。中間方可以是亞馬遜這類平台或市場，也可以是 Visa 這類處理支付的卡組織網路。
消費者應用通常負責處理資金或交易，並從中抽成。但當消費者是代表我們行事的 AI 時，情況會怎樣？目前有幾種新興標準正在尋找答案。
ChatGPT 擁有 7 億活躍用戶，他們都在嘗試通過 AI 獲取信息或服務。雖然我們尚未通過代理界面直接買賣商品，但已普遍用它來「發現」商品。無論是買跑鞋還是在埃爾卡拉法特找酒店，我都在用 AI 進行比價。如果能在同一個界面直接購買，無疑方便得多。這正是 OpenAI 與 Stripe 合作，推出自主代理商業協議（ACP）的目的。
Source: OpenAI
This is currently the most direct way for agents to handle funds: user-controlled. After the user places an order, ChatGPT sends the necessary information to the merchant's backend through ACP. The merchant then decides 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 ACP business as: you authorize an intern to spend a fixed budget, and you ultimately decide which product/service to choose, from which merchant to purchase, and complete the payment.
OpenAI and Stripe have ACP, while Google has introduced the Agent Payment Protocol (AP2). Before diving into AP2, let's take a step back. Google aims to solve the "interoperability" issue. Currently, AI agents operate in silos: Gemini does not interact with Claude, and ChatGPT doesn't know what's happening in Perplexity.
Ideally, when tasks become more complex and require collaboration, we hope that these agents can communicate in a common language. To achieve this, Google developed A2A (Agent-to-Agent Protocol), allowing different agents to communicate and coordinate.
But just being able to converse is not enough. Agents also need to be able to use tools, access APIs, and services. The Model Context Protocol (MCP) allows agents to use tools such as Google Calendar, Notion, Figma, and more.
Source: Level Up Coding
MCP defines a common language. As long as they both "speak" MCP, agents can use any tool without the need for additional customized code. The protocol was created by Anthropic, but the specification is open and is rapidly being adopted by various companies. The MCP server is essentially a translation layer that sits in front of a company's existing APIs, exposing services in a standardized format to any MCP-compatible agent.
Returning to AP2, it can be understood simply as follows: MCP gives agents the ability to access data, files, and tools; A2A gives them the voice to communicate with each other; and AP2 gives them a wallet to securely spend money.
All these protocols put the user at the center, with agents having limited spending permissions. This addresses distribution and process issues, but they still haven't resolved: what happens if the agen
