Industrial AI is moving toward greater autonomy, but Rob McGreevy of AVEVA says it still requires more testing, validation and human oversight. Current deployments range from advisory systems to narrowly autonomous operations, while companies and regulators work to define safety boundaries. Examples from refineries, food production, chemicals and energy show how AI, digital twins and robotics are already supporting industrial decision-making.
Human oversight remains central to industrial AI
“There’s a lot more work to do, more proof points, more testing, more validation,” McGreevy said of the continuing development of autonomous AI systems.
In the short term, he expects agentic AI to follow a path similar to other industrial technologies, including SCADA systems. “The ‘S’ in SCADA stands for surveillance, and surveillance means there is a human in between,” he said. McGreevy expects AI to operate primarily in an advisory role for now, with full autonomy limited to narrow, highly repetitive tasks.
The role of human operators is evolving. Workers who once specialized in a single process line or maintenance task can increasingly cover multiple areas. Rather than approving routine actions one by one, they spend more time defining acceptable operating limits. AI can identify patterns in the data it receives, making early error detection a key part of its role, while human knowledge and experience help interpret what the system observes and does.
Governance at both the enterprise and industry-wide levels also needs to adapt. Organizations must establish clear guardrails defining which decisions AI can make independently and how its actions will be monitored and audited. However, many organizations still lack comprehensive safety measures.
According to a report from AVEVA and IMD, 43% of organizations said AI had slightly changed or had not changed decision-making and accountability across ecosystem partners. Only 23% reported significant or fundamental changes. Just 27% said they shared data substantially or extensively with partners.
Regulators are beginning to establish rules as well. EU AI legislation classifies certain AI safety components used in critical infrastructure as high risk and imposes requirements for human oversight. Companies are also developing internal frameworks. AVEVA, for example, has a Responsible AI Governance Committee that has published principles for developing and deploying AI.
Agentic AI and living digital twins: Real-world case studies
Monitoring arrangements for semi-autonomous systems already in operation vary by deployment. In some cases, operators approve every action. In others, AI operates independently while humans monitor the system and define the boundaries within which it can act. McGreevy cited several examples among AVEVA customers:
- At Saudi Aramco’s Yanbu refinery, optimization algorithms adjust production in a closed loop and are monitored by operators as needed. At the Ras Tanura refinery, algorithms recommend changes and operators decide whether to implement them.
- Nestlé is using AI optimization at its Nesquik factory to adjust the industrial dryers used to make powdered beverage ingredients. In testing, the adjustment reduced energy use by 16% and improved quality and yield.
- At SCG Chemicals in Thailand, an AI-supported digital twin spanning engineering, operations and maintenance monitors asset health and flags potential failures. The company reports plant reliability of 99% and a nine-fold return on investment within six months.
- Rocky Mountain National Laboratory in the US has built a generative AI model for grid control rooms that provides large-scale monitoring. The model supports operators with renewable-energy forecasting and estimates of grid conditions.
Looking to the future
The lack of synchronization once seen between technical systems within individual plants also exists on a larger scale. “Manufacturing and critical infrastructure have operated as siled systems for the past 50 years,” McGreevy said. He believes emerging technologies could create opportunities for more coordinated, holistic operations.
At the enterprise level, this could involve integrating physical AI: systems that combine robotics and sensing technologies with AI-powered decision-making. Drones flying over pipelines and power lines, for example, could send visual inspection data to algorithms that identify gas leaks and other hazards. This approach could create a more accurate, interconnected flow of information while reducing the need for people to enter potentially dangerous environments.
“The convergence of robotics and software will lead to fewer incidents, fewer outages, and a more sustainable way of operating,” McGreevy said.
“The convergence of robotics and software will lead to fewer incidents, fewer outages, and a more sustainable way of operating.”
The broader vision extends beyond an individual plant or process to an entire industrial network—what McGreevy calls an “integrated value chain.”
“A living digital twin enables a digital representation of the entire value chain,” he said. “This allows companies to quickly respond to changes in supply chains, changes in demand, tariffs, and material availability.”
In a period of turbulent geopolitics and changing regulatory demands, McGreevy said, this resilience is “absolutely necessary.” “Without that, it becomes very difficult to run an effective business.”
Source: www.technologyreview.com












