Factories are beginning to use AI to do more than analyze equipment and recommend changes. Connected digital models can track live conditions, while agentic AI can act on selected insights. The challenge is giving these systems room to respond without losing human oversight of decisions that affect industrial operations.
From digital mirror to active system
A factory’s condition can change gradually. Wear and deterioration may first appear as small irregularities, so operators need timely information about equipment and processes to recognize when something is going wrong.
Earlier digital twins were mainly tools for simulation, observation, and what-if planning. They could represent operations, but they were not designed to analyze conditions in real time and proactively address detected issues. A living digital twin adds a dynamic representation of operations that can connect with control systems and help intervene in the physical plant.
AI supports this shift by helping digital models adapt as conditions change. In AVEVA chief product officer Rob McGreevy’s description, these models can connect systems that previously operated in isolation, giving teams more visibility into how a facility is working. That visibility can support maintenance, performance, and sustainability decisions.
Connecting data to decisions
A digital thread links engineering and design information with current operational data over an asset’s lifecycle. Sensors on IoT devices and industrial control equipment stream telemetry, software platforms make information accessible through user-friendly interfaces, and predictive or physics-based models can help estimate how equipment and processes might behave.
Operators can examine the resulting picture from different angles. They may ask whether emissions are on track, whether an asset is operating healthily, or whether production is meeting a target. These perspectives help teams focus on the conditions they need to manage, rather than treating the digital twin as a static replica.
With agentic AI, selected insights can lead to action. Small adjustments made in response to changing conditions could help keep a plant operating within defined limits. But the ability to act does not by itself establish what the correct limits should be, or which decisions a system should make independently.
Autonomy needs boundaries
Manufacturers are exploring the technology while weighing operational risk. A 2026 Manufacturing Leadership Council survey of manufacturers found that 66% said they are using or plan to use agentic AI in their operations, and 40% expect some degree of factory autonomy by 2030. In Deloitte’s 2025 smart manufacturing survey, 65% of executives ranked operational risk, including potential business disruption, among their top concerns.
McGreevy says more refinement, proof points, testing, and validation are needed. He expects industrial AI to remain largely advisory for the foreseeable future, with full autonomy limited to narrow, highly repetitive tasks. He compares the near-term role to SCADA systems: “The ‘S’ in SCADA means supervisory, and supervisory means there’s a human in the middle,” he says.
That arrangement changes the operator’s work. Instead of approving routine actions one at a time, workers may spend more time defining acceptable operating limits and checking whether AI behavior stays within them. Their experience matters because AI can reinforce patterns in the data it receives, making it important to identify when those patterns lead to missteps.
Companies also need governance that explains which decisions AI can make, and how its actions will be monitored and audited. In a study from AVEVA and IMD, 43% of organizations said AI had changed decision-making and accountability across ecosystem partners only slightly or not at all; 23% reported a significant or fundamental change. The findings point to a gap between deploying AI and adapting accountability around it.
Coordination beyond the factory floor
Oversight can take different forms: an operator might approve each action, or AI might act within boundaries that people set and supervise. The appropriate arrangement depends on the deployment, but the underlying need remains the same: operators must be able to understand what the system is doing and judge whether it is on track.
The EU AI Act classifies certain AI safety components used in critical infrastructure as high-risk and imposes requirements for human oversight. Companies are also developing internal approaches. AVEVA has a responsible AI governance board and published principles for how AI is developed and deployed.
Connected systems could eventually support coordination across an enterprise or industrial network. Drones inspecting pipelines and power lines can send visual data to algorithms that look for gas leaks and other hazards, reducing the need for people to enter potentially dangerous environments. At a broader scale, living digital twins could represent an integrated value chain, helping companies respond to changes in supply, demand, tariffs, and material availability.
Those possibilities depend on connecting information across systems while keeping responsibility clear. As industrial AI gains the ability to act, human judgment remains part of the control system: people define healthy operation, set limits, and watch for decisions that fall out of alignment.