Shanghai’s World Artificial Intelligence Conference wraps Monday, and China’s platform giants spent the final stretch selling something very specific: employees you subscribe to. On Sunday, Ant Digital Technologies — the enterprise-tech arm of Alibaba’s fintech affiliate Ant Group — unveiled Agentar 2.0, a platform it calls a “commercial AI agent super factory.” The pitch, per South China Morning Post and Chinese tech media: 200 pre-configured “digital expert” templates — job-level agents, not chat windows — plus hundreds of subscribable, out-of-the-box agent tools, so companies deploy autonomous workers without building anything from scratch.
It wasn’t a solo act. Across the four-day conference, Tencent, Alibaba, and Baidu all launched or showcased enterprise-agent offerings aimed at wiring AI into day-to-day business operations — the software half of the same flex China’s robot makers performed on the show floor Saturday. Ant Group CEO Cyril Han said the goal is agents that restructure entire workflows rather than speed up individual tasks: “This shifts the focus from individual efficiency to a wholesale upgrade of the organisation’s collective productivity.”
The factory has a track record to point at. Agentar — upgraded Sunday from a version launched last year — has already put more than 300 agents into production across Chinese banks, brokerages, and insurers, and Ant Digital says the push now extends into energy, transportation, and retail. In a heavily regulated financial sector, “deployed” carries more weight than “demoed.”
Templates are the moat, not the model
Notice what’s missing from the pitch: benchmark charts. While U.S. labs fight over frontier-model rankings — and build secret projects to out-reason each other — China’s giants are competing on distribution: prebuilt roles, subscription pricing, compliance-ready deployment inside industries they already bank, insure, and host. It’s the SaaS playbook applied to labor. The usual caveats apply — every number here is a company claim, “deployed” can mean anything from mission-critical to pilot-in-a-sandbox, and nobody publishes agent failure rates. But the strategic direction is unambiguous, and it’s the same one U.S. firms are backing into from the other side: agents doing defined jobs, at scale, under supervision.
Our take: The agent war just split into two races. America’s labs are selling smarter brains; China’s platforms are selling filled job reqs. If agents become catalog products — picked by role, billed like seats — the moat shifts from model IQ to workflow lock-in, and the platform that owns the enterprise relationship wins even with a second-best model. That’s bad news for every startup pitching “agents for enterprise” as a standalone product, in both countries. And watch the geography: Ant already runs fintech rails across Southeast Asia. The super factory is domestic today; the playbook travels in a shipping container marked SaaS.
What to watch
- Real adoption numbers. Renewals, seats, and production incident disclosures — not “deployed” counts. The first enterprise that publicly quantifies agent ROI (or agent damage) resets the debate.
- The U.S. counter. Whether Microsoft, OpenAI, and Anthropic lean harder into prebuilt role templates for enterprises — the early signs are already there — or keep betting that a smarter general model beats a configured specific one.
- Export routes. Ant Digital’s overseas push, starting with Southeast Asian financial institutions where Ant’s rails already run. An agent platform crossing borders will meet the same security scrutiny as China’s exported robots.