Building a safer path to autonomous industrial AI

Industrial automation is entering a new phase as foundation models, physical AI, and agentic systems move beyond traditional predictive analytics into complex physical environments. In a recent episode of the Business Lab podcast hosted by Megan Tatum, MIT Technology Review's report highlighted how industrial software provider AVEVA is navigating this shift toward autonomous decision-making. Industrial sector AI adoption has surged by nearly 78% over the past two years, creating a step-function acceleration in how plants, power systems, and mining facilities operate. Unlike software-only automation, industrial AI directly controls physical machinery where unpredictable model behavior carries severe risks to critical infrastructure and worker safety.

This article examines how industrial operators are combining disparate data sets, physical robotics, and governance frameworks to balance autonomous capabilities with human oversight. Chief technologist Arti Garg outlined AVEVA's operational strategy for integrating generative models and physical AI into mission-critical workflows without compromising safety or system reliability. The discussion provides a technical and operational breakdown for engineering leaders, plant managers, and software architects tasked with deploying autonomous AI across high-consequence industrial facilities.

Building a safer path to autonomous: what actually changed

Industrial AI has evolved from narrow, rule-based algorithms into foundation models capable of automating complex multi-step workflows. While AVEVA has developed industrial AI solutions for more than 20 years, the recent explosion in physical AI and agentic systems enables automated software execution alongside physical robotic hardware. Industrial sector adoption has jumped 78% in the past two years, moving past initial industry hesitancy surrounding unpredictable outputs. These general-purpose foundation models allow domain experts to build applications directly through AI-assisted coding, drastically expanding who can create custom automation tools.

The transition to physical AI allows autonomous robots and drones to inspect high-risk machinery and collect operational data directly within hazardous environments. This shift reduces the need for human workers to physically enter dangerous plant zones, transforming routine maintenance into remote oversight operations. However, because foundation models and agentic software are inherently less predictable than traditional deterministic systems, industrial organizations are forced to rethink traditional safety validation and system architecture before deploying autonomous capabilities at scale.

Building a safer path to autonomous: how it works

Modern industrial plants generate vast amounts of data across telemetry sensors, equipment service logs, engineering schematics, and maintenance documentation. AVEVA's architecture uses graph databases and AI algorithms to instantly match and correlate these disparate data sources. For example, telemetry coming off a pump or mixer can be matched immediately with its maintenance history and original design blueprints. Instead of an operator manually researching component connections across multiple disconnected databases, an AI assistant on a mobile tablet fetches and correlates the relevant operational data in near real time to diagnose active failures.

In physical inspection workflows, autonomous robots and drones navigate plant floors using onboard computing or direct wireless links to human operators. These machines gather visual and thermal telemetry from machinery operating in extreme or hazardous locations. The collected data feeds directly into correlation engines, allowing diagnostic software to flag abnormalities before equipment breakdowns occur. By combining graph-based data integration with physical AI hardware, the platform converts unstructured documentation and live sensor feeds into structured, actionable diagnostics for maintenance teams.

Building a safer path to autonomous: why it matters now

Deploying autonomous systems into critical infrastructure carries minimal margin for error, as unverified AI decisions can trigger physical equipment failures or safety incidents. AVEVA addresses this challenge by establishing a governance framework built around security, efficiency, human safety, and strict operational guardrails. Under this architecture, AI models are designed to augment human decision-makers rather than replace them in critical operational loops. Guardrails explicitly define which routine actions an automated agent may execute independently and which high-consequence decisions require mandatory human approval.

Environmental sustainability represents another critical driver for responsible industrial AI deployment. As renewable energy generation integrates into global power grids, AI systems manage the resulting load complexity across distribution networks. Concurrently, industry leaders are addressing the environmental footprint of AI computing itself. Garg is participating in an IEEE working group focused on standardizing methodology to measure AI's resource consumption across electricity, water, carbon emissions, and total energy usage.

Building a safer path to autonomous: who is affected

Plant managers, reliability engineers, and field operators across manufacturing, power generation, and mining sectors face immediate operational changes from autonomous systems. Experienced plant workers transition from manual diagnostic research and dangerous physical inspections into supervisory roles. Through AI-assisted coding tools, domain experts without traditional computer science backgrounds can build specialized software tools tailored to their specific operational challenges, changing the internal software development lifecycle within industrial facilities.

Executive leadership and safety officers are equally impacted as governance and liability frameworks adjust to non-deterministic software. Organizations must restructure business processes to enforce human-in-the-loop guardrails while training frontline personnel to operate alongside autonomous drones and mobile robots. By shifting routine monitoring to physical AI hardware, companies reduce occupational hazards for field technicians while demanding higher technical literacy in data analysis and AI supervision.

Building a safer path to autonomous: what to watch

The next phase of industrial automation will test whether organizations can adapt business processes to match rapid technical advancements in agentic and physical AI. Key performance metrics will focus on whether autonomous robots and real-time data correlation deliver measurable improvements in plant productivity and worker safety without introducing unplanned downtime. Industry adoption will depend heavily on the maturity of governance frameworks that successfully balance model autonomy with verifiable human oversight.

Watch for the formal release of standardized metrics from the IEEE working group regarding AI energy and resource consumption. Furthermore, track how industrial software providers implement guardrail mechanisms to maintain system reliability when deploying foundation models across mission-critical infrastructure. As autonomous drones, robots, and agentic coding tools expand across industrial sites, successful deployment will rely as much on process redesign and workforce training as on the underlying AI algorithms.