Illustration: AI-generated · Inside China AI
When the Agent Does the Shopping: China’s Agentic Commerce, Explained
Explainer · 20 November 2026
China’s e-commerce platforms are replacing the search bar with AI agents that compare, order and pay. Alibaba’s Qwen app can now search Taobao and Tmall and check out through Alipay in a single conversation; JD.com says almost 80 million people used its shopping assistant during this year’s Singles Day. What is actually live, what “autonomous shopping” means in practice, and why fashion is the hardest test.
Four things to know
- The search bar is being replaced. Alibaba, JD.com, ByteDance (Douyin/Doubao), Meituan and Tencent all run AI assistants that recommend, compare and send shoppers straight to checkout.
- Closed-loop agents are live, not a demo. Since May 2026 Alibaba’s Qwen app reaches more than four billion items on Taobao and Tmall and pays via Alipay (Reuters via Inside Retail Asia).
- Scale is real, but early. JD says its assistant Dongdong drew nearly 80 million users for Singles Day 2026, more than three times the year before (Reuters via The Star). Most use is still advice, not fully delegated buying.
- Whoever owns the agent owns the shelf. Platforms are fighting to be the default agent; Tencent has plugged JD into its Yuanbao assistant (TechNode). Brands now compete to appear in the agent’s answer.
What “autonomous shopping” actually is
Autonomous shopping means the customer states an intent and an AI agent executes the purchase: it searches, compares, picks, applies vouchers, pays with a delegated payment credential and tracks delivery, with the human only approving, or not even that.
In practice there are levels, much like driving automation. China has reached level 3 at scale. Level 4 exists in pilots and for low-risk repeat purchases such as food and drinks, not yet for fashion.
| Level | What the AI does | Example in China |
|---|---|---|
| 1 Assist | Answers questions, recommends products | Taobao and JD shopping assistants, Doubao in Douyin |
| 2 Curate | Compares, builds a basket, tracks prices, virtual try-on | Qwen assistant in Taobao (30-day price tracking, try-on) |
| 3 Transact | Places the order and pays after one confirmation | Qwen app → Taobao Instant Commerce → Alipay |
| 4 Delegate | Buys on its own within rules and budget | Pilots; voice ordering in Meituan’s AI food app |
The enabler is not the model but the stack: catalogue, payment and logistics in one ecosystem. Alibaba owns Taobao, Alipay and Cainiao; Meituan owns its delivery fleet. That is why closed-loop agents arrived in China first, while Western players are working through open protocols between separate companies.
Who is building what
- Alibaba (Qwen). January 2026: agentic features in the Qwen app (“I want bubble tea” → search, compare, buy). May 2026: Qwen connected to Taobao and Tmall with Alipay checkout; a live demo ordered 40 bubble teas, including loyalty discounts (The Next Web).
- JD.com. Shopping assistant Dongdong; around 80 million users at Singles Day 2026. Tencent’s Yuanbao assistant integrated JD’s agent mini-programs for recommendation, ordering, logistics and after-sales.
- ByteDance. Its Doubao chatbot has been linked to Douyin e-commerce since Double 11 2025 (Caixin).
- Meituan. A voice-command AI app for food ordering, launched in September 2025 against JD and Alibaba in instant retail (Yicai Global).
A caution on numbers: user counts are company figures and are defined differently (“users”, “AI shopping experiences”, “advice sessions”). We found no independent data on how many orders are placed entirely by agents.
Case study: Taobao puts the fitting room into the agent
Fashion is where agentic shopping hits its limits. An agent can reorder bubble tea, but it cannot know whether a jacket fits or looks right. Alibaba’s answer is to bring try-on into the agent.
- Virtual try-on at scale. Alibaba researchers describe their try-on model Tstars-Tryon 1.0 as deployed “at an industrial scale” in the Taobao app, serving millions of users with tens of millions of requests (paper).
- Try-on inside the conversation. The Qwen-powered assistant launched inside Taobao in May 2026 combines virtual try-on with 30-day price tracking, so advice, fitting and purchase happen in one flow (Alibaba).
- The business case is returns. Industry estimates put online apparel returns at around 30 per cent in key markets, mostly driven by fit (Style3D, a try-on vendor). Every agent-placed order that comes back damages trust in the agent.
What we do not know: Alibaba has not published how try-on affects conversion or return rates on Taobao. Figures on fewer returns come mostly from try-on vendors or Western retailers, not from independent Chinese data.
Why it matters for Europe
- Agent visibility is the new SEO. If the agent picks, product data, sizes, return terms and reviews must be machine-readable and trustworthy.
- Own the agent or live inside it. China shows the split between platforms with their own assistant and brands that sit inside a super-app’s agent. Europe is heading for the same choice with ChatGPT, Gemini and others.
- Payment is the bottleneck. China solved delegated payment inside one wallet. In Europe, PSD2 strong customer authentication and the AI Act shape how far fully delegated buying can go.
- Fashion is harder than bubble tea. Fit, taste and returns keep humans in the loop longer. Size data and try-on are where the competition will be decided.
The open question for Europe: will its retailers build their own shopping agents, or end up as suppliers to someone else’s?



No responses yet