China’s AI Agents Aim to Replace Attention With Intention

A paradigm shift: from scrolling to solving

China’s consumer internet is pivoting from the attention economy to what thinkers once called the intention economy: users state goals, systems complete tasks. The idea, popularized by Doc Searls in 2012, lacked practical rails—until large language models and AI agents arrived. Now the battleground is not “which app do you open” but “who hears your first request.” Who owns your first utterance? In this agent-led world, time-on-app gives way to task completion and fees, flipping a two-decade-old playbook.

Alibaba, Tencent, ByteDance race to capture the “first utterance”

Alibaba (阿里巴巴) is wiring its Qwen (千问) model into Taobao (淘宝, marketplace), Alipay (支付宝, payments), Fliggy (飞猪, travel), and Amap/AutoNavi (高德地图, maps) so a single prompt can order food, book trains, or plan routes end-to-end. It has been reported that a “one-cent milk tea” promotion drove over 10 million orders in nine hours—evidence that execution, not exposure, is the sticky point of the new funnel. The catch? If Qwen’s intent understanding lags top-tier models, a superior commerce stack may still miss the handoff. Tencent (腾讯) leans on WeChat (微信): many intents are inherently social—scheduling dinners, sending group red packets, nudging colleagues—plus Mini Programs and WeChat Pay form a ready-made dispatch hub. Yet Tencent’s “do less, disturb less” product ethos may clash with an agent that must proactively act. ByteDance (字节跳动) is pushing Doubao (豆包), which reportedly surpassed 100 million daily active users, and is partnering with ZTE (中兴通讯) on an “AI phone” to intercept intent at the operating system—before any app opens.

The new black box: who does the agent really work for?

Agents can bypass ad-driven recommendation feeds, but their choices remain opaque. Why recommend Merchant A over B? Possibilities include higher commissions (echoes of Baidu’s 百度 paid-ranking era), cleaner APIs that are easier for agents to call, or simple corporate affiliation—an Alibaba agent tilting to Alibaba merchants, a Tencent agent to the WeChat ecosystem. Users trade one form of platform steering for another, with even less visibility because delegation happens off-screen. Is power really returning to the user, or just migrating into a subtler box?

Policy and geopolitics shape the stakes

China’s regulators have pushed for algorithm filings and content governance, which could extend to agent transparency and conflicts-of-interest disclosures. At the same time, U.S. export controls on advanced semiconductors constrain the compute available for training state-of-the-art models, pressuring Chinese platforms to optimize inference and vertically integrate services to win on execution rather than pure model supremacy. The outcome is likely more winner-takes-most than the attention era: users might browse multiple feeds, but they will not ask two agents to book the same ticket. In the intention economy, whoever secures the first request controls the rest of the chain.