On 21 October the Ministry of Commerce, with eight other departments, issued an opinion on accelerating technology-driven commerce (科技兴商). The text was published in full by the Cyberspace Administration of China. It names the technologies it wants applied inside retail and trade: artificial intelligence, big data, IoT, cloud, blockchain, digital twins and embodied intelligence.
The concrete AI measures, in the document's own ordering: extend the "AI + manufacturing" campaign so that phones, computers, home devices, wearables and robots get smarter; use industrial large models and edge intelligence in flexible manufacturing and made-to-order production; build a new category of "commercial agents" (商业智能体) covering shopping guidance, store operations and logistics; put embodied intelligence into home services and trade logistics; extend the national standard series on grading the intelligence level of AI terminals; and build commerce-specific high-quality datasets and trusted data spaces.
On governance, the document asks for "prudent" tiered regulation of commercial agents by scenario, a negative list for data use, a monitoring system for data-security risk, and regulatory sandboxes that allow stress tests of new models within controlled limits. It also points to the AI start-up support programme and inclusive compute for AI ventures in the trade sector.
Our takeThe important line is not the use cases; it is "tiered governance of commercial agents", written before a public method for measuring agent behaviour exists. A grading scheme means something only if each tier maps to a testable capability and a stated failure mode, and the only standard series named here covers terminal intelligence levels, not agent reliability in a store. Sandboxes suit that gap, but results compare only if scenarios and failure criteria are published. Watch whether the negative list arrives with methods attached.