ibaba is actually deploying here. Taobao, alongside sister platform Tmall, is the world’s largest e-commerce marketplace by gross merchandise value. Last year, Taobao alone handled an estimated US$723 billion in gross merchandise value. Tmall added a further $682 billion. Together, they dwarf Amazon’s third-party marketplace by a factor approaching three. Alibaba’s China e-commerce business generates around 140 billion yuan in revenue per quarter. Dropping an AI agent into that environment is a different proposition from, say, OpenAI building shopping features through third-party plug-ins, or Amazon layering AI recommendations onto a search page. Qwen is not a chatbot grafted onto a retailer but a large language model trained, in part, on more than two decades of Alibaba’s own transaction data, merchant operations and consumer behaviour. The knowledge base is proprietary and, critically, embedded at the transaction layer rather than above it. Most global AI platforms pursuing commerce must begin trust-building and personalisation from scratch, relying on external integrations that introduce friction at every handoff. Alibaba has already resolved the hardest problem in AI-driven retail: it owns the data, the catalogue, the payment rails through the Alipay ecosystem, and the logistics network through Cainiao. Qwen is the last piece, the conversational interface that ties them together. The shopping companion  The mechanics of the new experience illuminate how different this model is from conventional e-commerce. A user asking what to buy for a friend’s birthday does not receive a list of keyword results. Qwen asks follow-up questions about budget and taste, surfaces relevant options, and completes the order in the same conversation. Someone redecorating a small apartment can describe the problem. The assistant returns a curated product bundle with attached styling notes. The Taobao-embedded Qwen Shopping Assistant, accessed through the app’s messages tab, goes further. It fields detailed product queries from camping gear suited to young children to electric toothbrushes safe for sensitive gums, and delivers side-by-side comparisons informed by Taobao’s billions of customer reviews. A virtual try-on feature uses Alibaba’s multimodal model to simulate how a garment looks on a user’s photo, capturing fabric texture and drape. The company plans to extend this to footwear and accessories. On the financial side, the assistant aggregates platform discounts during major shopping events, including the upcoming 618 mid-year sale, and recommends the most advantageous coupon combinations at checkout. A 30-day price tracker allows a user to set a target price for a specific item, and when the threshold is met, the assistant automatically places the order. Each of these features is interesting individually. Together, they describe something structurally significant: a system that does not wait to be told what to buy, but monitors conditions and acts on behalf of the consumer. That is the architecture of agentic commerce, and it is a meaningful departure from the passive recommendation engines that have defined e-commerce for the past two decades. “Qwen is by far the most effective at this compared to other LLMs,” said Jacob Cooke, co-founder and CEO at WPIC Marketing + Technologies. “Qwen has been trained on Alibaba transaction data and now the entire product catalogue, so it’s far more adept than other models at understanding what consumers want.” “I’ve been trying agentic commerce on platforms outside of China too, and even on non-Chinese platforms, Qwen is the most effective model at browsing online product catalogues and giving recommendations.” Competition heats up  Alibaba is not alone in pursuing this direction. ByteDance upgraded its Doubao AI chatbot late last year to handle autonomous tasks, including ticket bookings through integration with Douyin’s e-commerce features. Tencent’s Yuanbao and JD.com are each pushing AI deeper into their consumer surfaces. China’s technology sector is entering what analysts are now calling the agentic commerce race, a phase in which the competitive advantage shifts from having the best product catalogue to having the most capable AI that can navigate it on a user’s behalf.Alibaba has, for years, held the largest commercial dataset in China. It is now converting that dataset into an active agent. The consequences for how Chinese consumers shop, how merchants sell, and how rivals compete will take years to fully resolve. But the direction is clear. “In the next few years, I think we’ll see agentic commerce emerge as the most widespread consumer application of agentic AI, and it’s clear that Qwen is winning the race,” Cooke concluded. Further reading: Billions in red envelopes: Inside the battle for China’s AI super-app future.