SHANGHAI, March 23 (Reuters) – Alibaba strengthened its presence in the global race for agentic artificial intelligence after its international trade division launched Accio Work, an ‘AI task force’ that the company says can autonomously execute complex business operations for small and medium-sized businesses.
The launch comes amid a boom in China around agentic AI, triggered by OpenClaw, which has consumers, from students to retirees, rushing to join the trend called ‘lobster raising’, alluding to the tool’s lobster symbol.
Accio Work stands in contrast to the consumer frenzy, as the company claims it implements cross-functional AI teams that require no coding or configuration.
“We distinguish ourselves by being a specialized B2B tool rather than a generalist platform,” said Alibaba International Vice President Kuo Zhang. ‘We draw a very clear line on high-risk operations…any action involving financial transactions, executing payments, or accessing private files requires explicit, granular permission from the user.’
The launch comes less than a week after another division of Alibaba introduced Wukong, an enterprise-focused agentic AI platform that can coordinate multiple AI agents to perform complex business tasks, including document editing, spreadsheet updates, meeting transcription and search, in a single interface.
Alibaba also said last week that it will separate its AI business from its cloud computing arm. The newly formed Alibaba Token Hub business group, led by chief executive Eddie Wu, is the clearest indication yet that the company is shifting its focus to digital assistants that use far more tokens – units of data – than traditional question-and-answer chatbots.
Continues after advertising
Zhang said the global push to define agentic AI brings inherent risks that can only be mitigated with controlled and specialized models that balance automation and security.
‘We believe the biggest risk is in the use of horizontal and generalist models for vertical business tasks. By focusing on specialized B2B agents and implementing AI alongside layers of human approval, we can deliver the benefits of an autonomous workforce without the traditional risks associated with unconstrained AI,’ he said.
