Living digital twins put AI on factory controls; governance is lagging

Digital twins, software models of a physical asset or process, used to work like rearview mirrors: they simulated a plant and suggested what might happen. The pitch in this MIT Technology Review piece is a ‘living’ twin that keeps pace with real conditions and can act through control systems. Newer twins link design and engineering records to live operational data through a so-called digital thread, enabled by cheaper sensors and better software platforms. That moves AI from advising engineers to adjusting refineries, dryers and grids.

Demand is real. A 2026 Manufacturing Leadership Council survey cited in the piece found 66% of manufacturers using or planning to use agentic AI, and 40% expecting some factory autonomy by 2030. A 2025 Deloitte survey found 65% of executives ranked operational risk among their top concerns. Appetite and anxiety are running side by side.

Plants already run on sensors, control loops and supervisory systems, so a model making small, constant adjustments is an evolution rather than a leap. The examples are concrete: a Nestlé dryer tuning test that cut energy use by 16% while improving quality and yield, and SCG Chemicals reporting 99% plant reliability with a twin watching asset health. The piece is also candid about the middle ground. AVEVA’s product chief points to SCADA, where the S stands for supervisory and a human stays in the loop. Aramco shows both modes at once: one refinery runs closed-loop optimization under oversight, while at another algorithms only recommend and operators decide. Autonomy where tasks are narrow and repetitive, advice everywhere else, is the right shape for the next few years, echoing calls for an AI ‘emergency brake’ (Nadella’s brake).

The piece carries a vendor’s voice and vendor case studies, and the figures are reported by the companies themselves; a claimed ninefold return in six months needs independent audit before anyone treats it as normal. The bigger worry is governance. The AVEVA and IMD study found only 23% of organizations saw a significant change in decision-making and accountability across partners, while 43% saw little or none. Plants are wiring in agents faster than they are rewriting who answers when one gets it wrong.

If that gap persists, the likely failure is not a dramatic robot mishap but quiet drift: a model reinforcing patterns in its own data, operators approving less and noticing less, and small errors compounding across a value chain that used to be siloed. Integration that makes plants resilient also lets failures travel. In Europe, the AI Act already treats certain AI safety components in critical infrastructure as high-risk, with human-oversight requirements. More in our AI coverage.