AI agents face trust issues in high-risk industrial sectors

Reporting by the South China Morning Post (SCMP) from the IDC CIO Summit in Shenzhen describes a shift from chatbots toward agentic AI, systems capable of independent workflow execution. SCMP quotes Liu Xiangyang, chief information security officer of Midea Group, saying "Upper-layer software, which includes consumer-facing apps and internal management systems, will be 'entirely replaced by agents.'" SCMP also quotes Du Yanze, senior research manager at IDC, saying "In the future, 90 per cent of an AI agent's value will come from industrial expertise," and cites Du's example that agents could reduce a manual supply-chain order from two hours to several minutes. SCMP reports experts warned that critical verticals such as healthcare and aerospace may be too "high risk" for autonomous agents. SCMP further reports Beijing's "AI Plus" strategy sets adoption targets of over 70% by 2027 and more than 90% by 2030, and notes dependencies on foreign tools in advanced manufacturing software.

Agentic systems shift the bottleneck from generic language capability to domain-specific knowledge, labeled in the reporting as industrial expertise. Industry-pattern observations show this raises requirements for high-quality, curated industrial datasets, explainability tied to operational metrics, and integration with control-plane software such as MES and SCADA. Verification and deterministic safety checks become harder where training data for rare failure modes is sparse.

Observers following China's industrial strategy will note the tension SCMP reports between ambitious national adoption targets and existing software gaps in advanced manufacturing. For practitioners, that gap typically implies extended pilots, stronger change-control, and more rigorous validation before broad deployment in safety-critical workflows.

Observers should follow: regulatory guidance for agent deployment in regulated sectors; availability of domain-specific benchmark datasets; pilot results published by major manufacturers; vendor disclosures on verification and explainability; and integration case studies connecting agents to operational control systems.