China’s enterprise agents surge but face 70% success rate, says The Paper

The Paper’s snapshot of China’s “agent江湖” captures a transitional moment: technical capabilities for multi-step, tool-using agents are exploding, but organizations are struggling to turn pilots into robust systems. Executives quote 18‑month deployment cycles instead of the 3‑month timelines hoped for a year ago, and a 70% success rate that underperforms conventional IT. That gap between hype and durable value is exactly where enterprise AI strategies will be made or broken.

From an AGI lens, this matters because enterprise deployments are where abstract capabilities meet messy constraints—legacy systems, human incentives, and shifting requirements. The article’s anecdotes about teams debating whether to scrap last year’s agent architecture for something like OpenClaw show how quickly the ground is moving. Successful players will be those who treat agents as long-lived products that need governance, monitoring and iteration, not one-off PoCs.

It also illustrates that state-owned and heavily regulated sectors in China are leaning in, albeit cautiously. If agents become the default interface to internal knowledge and workflows in these massive organizations, they will generate both demand for stronger reasoning models and pressure to formalize standards around robustness and safety. That institutionalization, even if uneven, nudges the ecosystem toward the kinds of infrastructure you’d need to safely operate AGI-class systems.