China’s manufacturing and industrial sectors are aggressively pursuing the integration of artificial intelligence to boost efficiency and maintain global competitiveness. Bolstered by strong state support and ambitious national targets, the focus is rapidly shifting from conversational chatbots to “agentic AI,” systems capable of independently executing complex workflows. However, industry experts are sounding the alarm that critical, high-risk vertical markets may not be ready for the autonomous revolution.
Speaking at the International Data Corporation (IDC) CIO Summit in Shenzhen on May 16, 2026, technology leaders highlighted the immense potential and significant hurdles in deploying AI agents in industrial settings. The consensus among experts is that while agentic AI represents the future of enterprise software, the transition requires a fundamental shift in how AI models are trained, verified, and trusted.
According to reporting by the South China Morning Post, summarized by Let’s Data Science, the vision for agentic AI is sweeping. Liu Xiangyang, chief information security officer of Midea Group, one of China’s largest appliance manufacturers, predicted a total transformation of the software stack. “Upper-layer software, which includes consumer-facing apps and internal management systems, will be ‘entirely replaced by agents,'” Liu stated at the summit.
The Value of Industrial Expertise
The shift toward agentic systems moves the bottleneck of AI development from generic language capabilities to highly specific domain knowledge. Du Yanze, a senior research manager at IDC, emphasized this point during the summit. “In the future, 90 per cent of an AI agent’s value will come from industrial expertise,” Du noted.
When properly trained on domain-specific data, the efficiency gains can be staggering. Du provided an example where AI agents could reduce the time required for manual supply-chain order processing from two hours to just several minutes. This level of automation is exactly what Beijing envisions with its “AI Plus” strategy, which sets aggressive adoption targets of over 70 percent across industrial sectors by 2027, and more than 90 percent by 2030.
To support this rapid deployment, the government is actively building the necessary infrastructure, including specialized testing facilities like the national pilot base for embodied AI in Hangzhou.
The Trust Deficit in High-Risk Sectors
Despite the enthusiasm, experts at the IDC summit warned that critical vertical markets, specifically identifying healthcare and aerospace, may be too “high risk” for the immediate shift toward autonomous agents.
The core issue is trust and verification. Large language models (LLMs), which form the cognitive engine of many AI agents, often struggle in professional domains due to a lack of highly curated industrial data and specialized knowledge during their training phase. In a consumer chatbot, a hallucination or logical error is an inconvenience; in aerospace manufacturing or healthcare diagnostics, it can be catastrophic.
Deploying agentic AI in these sectors requires rigorous safety checks, explainability tied to operational metrics, and seamless integration with existing control-plane software, such as Manufacturing Execution Systems (MES) and Supervisory Control and Data Acquisition (SCADA) systems. Verification becomes exceptionally difficult because training data for rare, high-consequence failure modes is inherently sparse.
This trust deficit is further complicated by existing gaps in China’s industrial software ecosystem. The SCMP report noted that China’s industrial software stacks still rely heavily on foreign tools in advanced manufacturing. Integrating domestic AI agents with foreign legacy systems adds another layer of complexity and potential vulnerability to the deployment process.
Regulatory and Technical Responses
The challenges highlighted at the IDC summit are not going unnoticed by regulators. Recognizing the unique risks posed by autonomous systems, Beijing has already begun drafting frameworks to manage the transition. Recent draft guidelines for AI agent security and governance emphasize the need for strict human oversight, robust testing protocols, and clear liability structures before agents can be deployed in critical infrastructure.
For practitioners and enterprise CIOs, the message from Shenzhen is clear: the era of agentic AI is arriving, but it will not be a simple plug-and-play upgrade. Bridging the gap between ambitious national adoption targets and the reality of safety-critical workflows will require extended pilot programs, stronger change-control mechanisms, and the development of massive, high-quality industrial datasets.
As companies like Midea push forward with agent-driven architectures, the true test of China’s industrial AI revolution will not be how quickly it can deploy agents, but how reliably those agents can perform when the stakes are highest. The race is on, but in the high-risk sectors of the economy, trust must be earned before autonomy can be granted.