Beware of AI in E-Commerce, Alibaba Warns

Alibaba.com President Zhang Kuo said merchants making money in e-commerce should be wary of AI, citing tests showing that the most advanced frontier models completed just over 61 percent of real e-commerce tasks without human intervention.

Speaking on a Chinese podcast in October 2026, Zhang said his team compiled 107 e-commerce tasks from over a million real operational records from Alibaba.com and Accio. Tested under fully unsupervised conditions, the most advanced models passed barely more than 61 percent – far from the "99-point leaderboards" touted at every model launch.

Zhang's conclusion: high scores in coding, math, or general agent capabilities have almost no correlation with actually helping merchants make money in e-commerce. Business decisions on payments, procurement, budget allocation, and product positioning should remain in human hands, he said.

His team's agent, Accio Work, built on supply chain data from Alibaba.com and 1688, completed the same 107 tasks at roughly one-third the cost of Codex and Claude Code, using model routing based on Pareto optimality.