Qwen Office reaches 30 million users a month after launch, more than half from enterprise accounts.
The AMW Read
Qwen Office's enterprise-majority user base and vertically bundled model+IM+cloud stack update the agent player map and exemplify the emerging enterprise-harness distribution pattern also seen in Claudeforce and Microsoft 365 Copilot.
Named counterparties: Alibaba Group
Qwen Office reaches 30 million users a month after launch, more than half from enterprise accounts.
Alibaba's Qwen Office AI assistant hit 30 million users one month after launch, with enterprise accounts now over half the total. In early August Alibaba merged three separate agent products — QoderWork, MuleRun, and Wukong — into Qwen Office, embedding it inside DingTalk, which connects over 26 million enterprise organizations and 800 million users. On August 26 the team released Qwen3.8-Flash, tuned for multi-step planning and tool use, cutting average token consumption about 75% and roughly doubling task generation speed. It also open-sourced MyContext, an infrastructure layer resolving message timing, conversation boundaries, and conflicting information before an agent acts.
This tracks a shift the piece ties to OpenAI's August Enterprise Signals update — from assistance to delegation — and to the new Claude-Salesforce 'Claudeforce' partnership, pairing Claude's reasoning with Salesforce's CRM data as an enterprise harness. Alibaba already owns all three layers — model, DingTalk distribution, and Alibaba Cloud systems — without an outside partner. Microsoft disclosed over 30 million paid Microsoft 365 Copilot seats in late July, with net-new seats more than doubling quarter over quarter; a Gartner survey from September 1 found 85% of leaders plan to raise 2026 AI spending but only 22% have scaled it across business units.
Two cases show where value lands: Changan Automobile cut a two-day wiring-harness selection task to five minutes and vehicle-document checks from ten-plus minutes to one or two, using Qwen Office across R&D and production; a law firm converted one lawyer's decade of will-drafting expertise into a reusable 'Skill' other staff can call. The edge in enterprise agents is shifting from model benchmarks toward who controls the context layer and distribution — a stack incumbents with existing enterprise surfaces can assemble alone, while others need a partnership like Claudeforce to compete.
