本周一个被大多数人忽略的消息,可能是今年 AI 领域最重要的信号之一。
据多个信源透露,微信正在内测 A2A(Agent-to-Agent)协议的接入能力。简单说:未来在微信里,你的个人 AI Agent 可以直接和商家、服务商、机构的 Agent 对话,完成预订、购买、售后、挂号等各种事务——你不需要打开任何其他 App,不需要填写任何表单,甚至不需要亲自打字。
这个变化听起来不起眼,但它可能重新定义"互联网入口"这个概念。
A2A 是什么?为什么重要?
A2A(Agent-to-Agent Protocol)是 Google 在 2025 年初联合 50 多家企业推出的开放协议,目标是让不同平台、不同公司开发的 AI Agent 能够互相"说话"、互相协作。你可以把它理解为 Agent 世界的 HTTP——在 HTTP 出现之前,不同的计算机网络互不连通;HTTP 统一了通信标准之后,互联网才真正诞生。
A2A 要做的事情一样:让 Anthropic 的 Agent 能和 Google 的 Agent 协作,让企业的客服 Agent 能和用户的个人 Agent 对话,让不同生态系统里的 Agent 打破围墙。
Google 推动 A2A 是有战略意图的——它想成为 Agent 互联网的"标准制定者",就像它在 Android 生态做的那样。但 A2A 最大的变量不是 Google,而是微信。
微信的独特位置
Google 有 Android 和 Google Assistant,但它没有一个 13 亿月活、深度嵌入日常生活的超级 App。微信有。
想想看,微信已经承载了中国人日常生活的多少场景:聊天(社交)、公众号(内容)、小程序(服务)、支付(交易)、视频号(媒体)。从约饭到挂号到交水电费到办签证,几乎没有什么事情是微信做不了的。
但微信也有一个致命的体验问题:服务太多但入口太深。你想在微信里挂号,需要找到对应的医院小程序、注册、选择科室、选择时间、填写信息——每一步都有流失率,每一步都有用户因为"太麻烦"而放弃。小程序用完即走的理念是好的,但"找到小程序"本身就是一个门槛。
A2A 协议完美解决了这个问题。
Agent 社交场景:想象一下
接入 A2A 之后,微信里的体验会变成什么样?
场景一:订餐厅。你对你的微信 Agent 说"周六晚上七点,帮我在公司附近找一家人均 200 左右的日料,四个人"。你的 Agent 直接在微信里和大众点评/美团的餐厅 Agent 对话,查询空位、确认时间、用微信支付付定金——你只需要在最后确认一下就行。不需要打开大众点评,不需要筛选,不需要手动填信息。
场景二:电商售后。你买的耳机有问题,你的 Agent 直接和京东/淘宝的客服 Agent 对话,说明问题、提供订单号、要求退换货。如果是常见问题,两个 Agent 在几秒内就达成解决方案;如果是复杂问题,才升级到人类客服。不需要你自己打字描述问题、等客服回复、反复沟通。
场景三:日程协调。你想约三个朋友周末打球。你的 Agent 直接和三个朋友的 Agent 对话,比对四个人的日程、找合适的场地、确认时间——全部自动协商,最后给你一个确认通知。不需要在群里来来回回讨论"周六行不行""周日呢""三点怎么样"。
场景四:差旅安排。你的 Agent 和航司 Agent 对话查机票,和酒店 Agent 对话订房间,和滴滴 Agent 预约接送机,和公司报销 Agent 同步行程——整个差旅安排在几分钟内自动完成,你只需要说"帮我安排下周三去上海的出差"。
"互联网的第一次革命是'人找信息'(搜索),第二次是'信息找人'(推荐),第三次可能是'Agent 替人办事'。"
—— 某互联网产品老兵
为什么是微信,不是其他 App?
支付宝、美团、抖音都在做 AI Agent,但它们有一个共同的问题:它们都是"单领域"的 App。支付宝管钱,美团管本地生活,抖音管内容娱乐——你的 Agent 在这些 App 里只能做一类事情。
微信不同。微信是一个"通用通信层"——它不做所有事情,但它连接所有做事情的人(和服务)。公众号连接内容创作者,小程序连接服务商,微信支付连接交易,企业微信连接 B 端。微信天然就是一个"Agent 与 Agent 之间对话的场所"。
更关键的是社交关系链。Agent 之间的协作不是纯机器对机器的——它需要人的授权、人的社交关系、人的信任网络。你让你的 Agent 去和餐厅 Agent 订位,背后是你和朋友的社交关系、你对餐厅品牌的信任、你的支付习惯——这些数据全部在微信里。
对行业的影响
如果微信真的全面接入 A2A,几个深远的变化可能发生:
第一,小程序的形态会变。现在的小程序是"人操作的界面",未来的小程序可能进化为"Agent 调用的接口"。开发者不再需要做精美的 UI,而是需要提供标准化的 Agent 接口——就像从网站时代到 API 时代的转变一样。
第二,App 的价值会被重新定义。如果所有服务都可以通过 Agent 在微信里完成,用户为什么还要下载独立 App?这对所有非超级 App 都是巨大的威胁,但对微信来说是巨大的机会——它可能成为 Agent 时代的"操作系统"。
第三,新的商业生态会出现。Agent 之间的交易怎么分成?Agent 推荐服务的佣金模型是什么?Agent 的"身份认证"怎么做?这些都是全新的问题,也是全新的商业机会。
还远吗?
A2A 协议还在早期阶段,微信的接入也在内测阶段。安全问题(Agent 能不能替你花钱?)、隐私问题(Agent 之间的对话数据归谁?)、标准化问题(不同平台的 Agent 真的能顺畅对话吗?)都还没有完美答案。
但方向是清晰的:未来的互联网,不只是人在上网,Agent 也在上网。人负责决策和享受结果,Agent 负责沟通和执行。微信作为中国最大的"通信层",在这个转变中拥有得天独厚的优势。
十年前我们说"微信连接一切",那时候连接的是人和人、人和信息、人和服务。下一个十年,微信可能连接的是 Agent 和 Agent——而那才是真正的"万物互联"。
明天见。
A piece of news most people overlooked this week might be one of the most important AI signals of the year.
According to multiple sources, WeChat is internally testing integration with the A2A (Agent-to-Agent) protocol. Simply put: in the future, inside WeChat, your personal AI Agent will be able to chat directly with Agents of merchants, service providers, and institutions to complete bookings, purchases, after-sales, doctor's appointments, and all kinds of errands -- no other app to open, no forms to fill, you don't even need to type yourself.
This change sounds trivial, but it could redefine the very concept of the "internet entry point."
What Is A2A and Why Does It Matter?
A2A (Agent-to-Agent Protocol) is an open protocol launched by Google in early 2025 alongside 50+ companies, with the goal of enabling AI Agents developed on different platforms and by different companies to "talk" to and collaborate with each other. Think of it as HTTP for the Agent world -- before HTTP, different computer networks were disconnected; after HTTP unified communication standards, the internet as we know it was born.
A2A aims to do the same: let Anthropic's Agents collaborate with Google's Agents, let enterprise customer-service Agents talk to users' personal Agents, let Agents across different ecosystems break down walled gardens.
Google is pushing A2A with strategic intent -- it wants to be the "standard-setter" of the Agent internet, just as it did in the Android ecosystem. But the biggest wildcard for A2A isn't Google -- it's WeChat.
WeChat's Unique Position
Google has Android and Google Assistant, but it doesn't have a super app with 1.3 billion monthly active users deeply embedded in daily life. WeChat does.
Think about it: WeChat already carries so many scenarios of Chinese daily life -- chatting (social), Official Accounts (content), Mini Programs (services), Pay (transactions), Channels (media). From making dinner plans to booking doctor's appointments to paying utility bills to applying for visas, there's almost nothing WeChat can't do.
But WeChat also has one fatal UX problem: too many services buried too deep. Want to book a doctor's appointment in WeChat? You need to find the right hospital's Mini Program, register, select a department, pick a time, fill in your info -- every step has drop-off, every step has users giving up because "it's too much hassle." The "use and go" philosophy of Mini Programs is nice, but "finding the Mini Program" itself is a barrier.
The A2A protocol solves this problem perfectly.
Agent Social Scenarios: Imagine
After A2A integration, what does the WeChat experience become?
Scenario 1: Restaurant booking. You tell your WeChat Agent, "Saturday at 7 PM, find me a Japanese place near the office, around 200 RMB per person, for four people." Your Agent chats directly with the Dianping/Meituan restaurant Agent inside WeChat, checks availability, confirms the time, pays the deposit via WeChat Pay -- you just confirm at the end. No need to open Dianping, no filtering, no manually filling in info.
Scenario 2: E-commerce after-sales. The earphones you bought are defective. Your Agent chats directly with the JD/Taobao customer service Agent, explains the issue, provides the order number, and requests a return or exchange. If it's a common issue, the two Agents reach a resolution in seconds; only complex issues get escalated to a human. No typing out the problem yourself, waiting for CS replies, going back and forth.
Scenario 3: Scheduling coordination. You want to invite three friends to play ball on the weekend. Your Agent chats directly with your three friends' Agents, compares all four calendars, finds a suitable venue, confirms a time -- all negotiated automatically, with a confirmation notification sent to you at the end. No more group-chat back-and-forth of "does Saturday work" "what about Sunday" "how about 3 PM."
Scenario 4: Business travel arrangements. Your Agent chats with the airline Agent to check flights, the hotel Agent to book rooms, the Didi Agent to schedule airport transfers, and syncs the itinerary with your company's expense Agent -- the entire trip gets auto-arranged in minutes; all you say is "book me a business trip to Shanghai next Wednesday."
"The internet's first revolution was 'people finding information' (search), the second was 'information finding people' (recommendations), and the third might be 'Agents doing things for people.'"
-- An internet product veteran
Why WeChat, Not Other Apps?
Alipay, Meituan, and Douyin are all building AI Agents, but they share one problem: they're all "single-domain" apps. Alipay handles money, Meituan handles local services, Douyin handles content and entertainment -- your Agent can only do one category of things inside those apps.
WeChat is different. WeChat is a "universal communication layer" -- it doesn't do everything, but it connects everyone (and every service) that does things. Official Accounts connect content creators, Mini Programs connect service providers, WeChat Pay connects transactions, WeCom connects B2B. WeChat is naturally a "venue for Agents to converse with other Agents."
Even more critical is the social graph. Collaboration between Agents isn't purely machine-to-machine -- it requires human authorization, human social relationships, human trust networks. When you send your Agent to book a table with a restaurant Agent, behind that are your social relationships with friends, your trust in the restaurant brand, your payment habits -- all of that data lives inside WeChat.
Industry Impact
If WeChat truly integrates A2A across the board, several far-reaching changes could unfold:
First, the form factor of Mini Programs will change. Today's Mini Programs are "human-operated interfaces"; future Mini Programs might evolve into "Agent-callable APIs." Developers won't need to build polished UIs -- they'll need to provide standardized Agent interfaces, much like the shift from the website era to the API era.
Second, the value of standalone apps will be redefined. If every service can be completed through Agents inside WeChat, why would users download standalone apps? This is a massive threat to every non-super-app, but a massive opportunity for WeChat -- it could become the "operating system" of the Agent era.
Third, new business ecosystems will emerge. How are transactions between Agents rev-shared? What's the commission model for Agent-recommended services? How does Agent "identity authentication" work? These are all brand-new questions -- and brand-new business opportunities.
Is It Far Off?
The A2A protocol is still in its early stages, and WeChat's integration is in internal testing. Security questions (can Agents spend your money?), privacy questions (who owns the conversation data between Agents?), and standardization questions (can Agents from different platforms really converse smoothly?) don't have perfect answers yet.
But the direction is clear: the future internet isn't just people going online -- Agents are going online too. Humans make decisions and enjoy the results; Agents handle communication and execution. WeChat, as China's largest "communication layer," has a unique advantage in this transformation.
Ten years ago we said "WeChat connects everything" -- back then it connected people to people, people to information, people to services. Over the next decade, WeChat might connect Agents to Agents -- and that's when "internet of everything" becomes real.
See you tomorrow.
A2A protocol deep-dive - WeChat AI ecosystem roadmap - The Agent internet's TCP/IP moment - Super app Agent transformation