AI助手市场的第一次格局洗牌，正在发生。
两个信号几乎同时出现：一是微信内嵌AI助手「小微」悄悄开启灰度测试，用户不用下载新App，打开微信就能用，还能直接调用小程序；二是曾以3.36亿月活领跑的豆包，推出付费计划后两个月流失了约1173万用户，月活跌到3.24亿。
一边是为了盈利主动收缩规模，一边是靠生态流量悄悄铺路。两种选择，正在把AI助手市场推向一个没人预料到的方向。
生态分发的真正威力
你可以把生态分发理解成「借船出海」——不用自己造船找航线，直接跳上已经挤满人的巨轮。
微信的14亿月活，就是这艘巨轮。「小微」不需要用户额外下载，不需要注册账号，打开微信主界面就能找到入口。这种「零门槛触达」的优势，是任何独立AI助手都没法比的。
但生态分发的魔力不止于拉新。它能把AI助手和用户的日常行为绑在一起：想订外卖，直接让小微调起小程序下单；想整理聊天记录，它能自动生成摘要；甚至连发朋友圈、设置群公告都能代劳。这种「在熟悉的场景里解决问题」的体验，比打开一个陌生App要顺畅得多。
6月8日，微信已经正式向小程序开发者开放AI生态接入。自动模式零代码接入，开发模式基于MCP协议封装技能包。第一批内测名单里有美团、京东、滴滴、携程、肯德基——这些月活数千万甚至数亿的应用都在抢位置。
从「人找服务」到「服务找人」的转变，正在微信里真实发生。
豆包的困境：过早收费的代价
豆包的流失数据，其实折射了一个更本质的问题：AI产品什么时候收费才合适？
ChatGPT是在用户规模突破10亿、用户对AI的价值认知足够成熟后，才推出Plus订阅服务，而且至今保留免费基础版。这种「先做规模，再谈盈利」的节奏，更符合AI产品的成本结构和用户心理。
AI产品的付费从来不是「定个价收钱」这么简单。和传统软件近乎零的边际成本不同，AI助手每一次回答都要消耗算力——用户用得越多，平台的成本就越高。这让付费决策变得非常微妙：太早收费，用户还没建立依赖，就会跑；太晚收费，成本压力撑不住。
豆包的困境，恰恰踩中了「过早收费」的雷区。当用户还没完全习惯用AI解决日常问题，还没建立起「离不开」的依赖时，付费门槛就像一道墙，把那些只是「试试看」的用户挡在了外面。
但话说回来，独立AI助手也不是没有优势。专业的AI编程工具能帮开发者快速写代码，设计类AI能生成精准的视觉素材——这些垂直领域的AI工具，哪怕用户量不大，也能靠高付费率活得很好。
未来的AI助手市场，不会有绝对的赢家，只会有一批能精准满足用户细分需求的玩家。
明天见。
The first major shakeup in China's AI assistant market is happening right now.
Two signals arrived almost simultaneously: first, WeChat's embedded AI assistant 'Xiaowei' quietly entered gray testing — no new app download needed, it's right inside WeChat and can directly invoke mini-programs. Second, Doubao, once the leader with 336M monthly active users, lost roughly 11.73 million users in two months after launching paid plans, dropping to 324M MAU.
One company is actively shrinking its user base for profitability. The other is quietly building distribution through ecosystem traffic. Two choices are pushing the AI assistant market in a direction nobody quite anticipated.
The Real Power of Ecosystem Distribution
Think of ecosystem distribution as 'sailing on someone else's ship' — you don't build the boat or chart the course, you just jump onto a vessel already packed with people.
WeChat's 1.4 billion MAUs are that ship. Xiaowei doesn't require a separate download, doesn't need a new account. It's just there in the WeChat interface. That kind of zero-friction reach is impossible for any standalone AI assistant to match.
But the magic of ecosystem distribution goes beyond acquisition. It weaves the AI assistant into people's daily behavior: order takeout by having Xiaowei pull up the Meituan mini-program and place the order. Summarize long chat threads automatically. Even post to Moments or set group announcements — it can handle those too. Solving problems in familiar contexts is a fundamentally smoother experience than opening a separate app.
On June 8, WeChat officially opened AI ecosystem access to mini-program developers. Auto mode requires zero code. Developer mode wraps mini-program capabilities into skill packages via the MCP protocol. The first wave of beta testers includes Meituan, JD.com, Didi, Ctrip, KFC — apps with hundreds of millions of users, all fighting for position.
The shift from 'users searching for services' to 'AI serving users proactively' is actually happening inside WeChat.
Doubao's Dilemma: The Cost of Monetizing Too Early
Doubao's user loss numbers reflect a deeper question: when is the right time for an AI product to charge?
ChatGPT rolled out Plus only after surpassing 1 billion users and reaching sufficient maturity in user perception of AI's value — and it still keeps a free tier. That 'scale first, monetize later' rhythm aligns better with AI cost structures and user psychology.
Monetizing AI products isn't as simple as 'set a price and collect money.' Unlike traditional software with near-zero marginal cost, every AI assistant response consumes compute — the more users use it, the more it costs the platform. This makes the timing of paywalls delicate: charge too early, and users who haven't formed dependency leave. Charge too late, and cost pressure becomes unsustainable.
Doubao's predicament lands squarely in the 'too early' camp. When users haven't fully integrated AI into their daily problem-solving routines and haven't built that 'can't live without it' dependency, a paywall acts like a wall, filtering out everyone who was just 'trying it out.'
That said, standalone assistants aren't without advantages. Professional AI coding tools help developers ship faster. Design-focused AI generates precise visual assets. These vertical AI tools, even with smaller user bases, can thrive on high conversion rates.
The future AI assistant market won't have one absolute winner. It'll have a cohort of players that precisely serve specific user needs.
See you tomorrow.
