Alibaba.com president Kuo Zhang says the next stage of AI commerce is agents handling work such as supplier negotiation, product design and store management, rather than simply answering questions. His interview also offers a concrete test of the pitch: Alibaba.com says its Accio agent completed a 107-task benchmark for less than half the estimated cost of two rival coding tools.
Watch Desk analysis
What happened
In an interview with AI Magazine published on 16 October, Zhang described Accio Work as an enterprise platform for automating business operations. He said its third version was released this month, and that businesses are using its agents for tasks including market research and generating product design packs.
Zhang’s proposed formula for an effective agent is “Model × Harness × Context”: the model’s intelligence, the tools and guardrails that manage its work, and the business information it can use. He also described Accio Work as an open platform, with roughly 500 partners integrated, and said it can help merchants manage shops across platforms such as Shopify, Amazon and TikTok.
Alibaba.com’s benchmark compared Accio on 107 e-commerce tasks with OpenAI Codex and Anthropic Claude Code. The company put the estimated cost of the full task set at US$3.69 for Accio, US$9.27 for Codex and US$9.51 for Claude Code. Those are Alibaba.com’s reported figures, not an independent head-to-head assessment.
Why it matters
The interview makes agentic commerce more tangible than a prediction that software will soon “do things”. In Zhang’s account, that means moving through practical steps such as preparing a product design pack, finding suppliers and updating listings across different shops. If agents can reliably join up those jobs, businesses may spend less time on repetitive coordination and supplier back-and-forth.
The benchmark figures speak to another essential question: whether agents are affordable for smaller businesses, not just technically impressive. But a low estimated cost on a company-designed task set does not, by itself, show how well the system performs on other work or whether the savings hold in everyday use.
Our read
The useful idea here is Zhang’s emphasis on the plumbing around the model. A clever model without the right tools, business context or guardrails is an agent with a grand job title and no access badge.
Alibaba.com’s cost comparison is worth watching, but the real test is whether businesses can delegate useful tasks without trading away accuracy or control. Ask for evidence from real workflows, not just a tidy benchmark table.
What to watch
- Whether Alibaba.com publishes the benchmark tasks, methods and results in enough detail for others to reproduce them.
- What the third version of Accio Work can do, and which features are generally available.
- Whether businesses report reliable results from agents handling supplier, design and store-management tasks.
Discussion spark: Would you let an AI agent negotiate with suppliers or change your product listings, or should it stop short of taking action without human approval?
Sources and evidence
- Alibaba.com’s Kuo Zhang on The Future of Agentic Commerce – AI Magazine (16 October 2026, 07:39 UTC)
Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.