Forging a Safer Path Toward Autonomous Industrial AI

Meta description: AVEVA's industrial AI roadmap traces predictive analytics to physical and agentic systems, emphasizing safety, governance, and sustainable automation.

Tags: industrial AI, AVEVA, responsible AI, physical AI, agentic AI

Featured image: Abstract industrial control-room visualization showing connected sensors, robot inspection, and data graphs, with no people, logos, or text.

Industrial AI is entering a more consequential phase. The field has spent decades using predictive analytics and specialized applications to preempt equipment failures and improve efficiency. Now general-purpose foundation models, physical AI, and agentic AI are making it possible to automate more complex tasks across industrial environments. That shift matters because industrial AI is not confined to screens: it can interact directly with physical systems, where an unexpected decision can affect safety, reliability, and critical infrastructure.

The change is being discussed in a Business Lab episode produced in partnership with AVEVA, hosted by MIT Technology Review's Megan Tatum and featuring Arti Garg, chief technologist at AVEVA. Business Lab presents itself as a resource for leaders trying to understand technologies moving from research into the marketplace. In this case, the stakes are high-consequence operations with little margin for error.

Background: Two Decades of Industrial AI Before the New Wave

Industrial AI is not a new experiment at AVEVA. Garg describes the company's work in the space as spanning more than 20 years, with an early focus on helping industrial applications become smarter. The older generation of systems was largely specialized: models were selected for predictable behavior, deployed for defined tasks, and evaluated against known operational boundaries.

The change has come from the type of AI now available. Foundation models have made advanced AI capabilities accessible to a broader set of users. In the last couple of years, physical AI has accelerated robotics and autonomous systems, while agentic AI has extended automation to software tasks that once required human sequencing. The result is not a gradual upgrade but a step-change in what industrial operators can attempt to automate.

Development: Data, Robots, and Faster Diagnostics

The foundation of this transition is data. Industrial sites typically hold information across telemetry, service logs, engineering documents, design records, and maintenance histories. In the past, an operator diagnosing a pump or mixer might have had to search through those sources manually. Newer tools, including graph databases and AI-assisted correlation, can link disparate records much faster.

AVEVA’s example is practical: a worker could use a mobile tablet or iPad with AI to retrieve and connect relevant information in real time. A robot could move through a hazardous area, collect data with onboard computation or a quick link to an operator, and reduce the need for people to enter dangerous spaces. That is where the benefit becomes concrete—faster diagnosis, less exposure, and more operational visibility.

The adoption curve has also moved. Although industrial users have historically hesitated to adopt models without predictable outcomes, a study cited in the discussion suggests industrial-sector AI adoption rose almost 78% over the past two years. Garg frames this as an inflection point, but one that brings a difficult requirement: newer AI systems are more capable and less predictable than the best-fit models that have been used in industrial software for decades.

Latest Progress: Governance, Sustainability, and Human Oversight

As autonomy increases, governance becomes the limiting factor. AVEVA’s responsible-AI approach is built around a triple mandate: security, efficiency, and human safety and oversight, with environmental efficiency included in the efficiency pillar. Garg argues that AI should augment rather than replace people in critical decision loops. Automated systems need guardrails that define where they can act and where human supervisors remain responsible.

AVEVA describes this as a multi-layered governance model covering both internal use of AI and AI embedded in products. Underlying that model, human judgment, human responsibility, and human ethics remain central. The company frames this end-user focus as part of its core DNA, because industrial software is tied to real operations rather than purely digital outputs.

Sustainability is part of the same equation. AI can help manage complex power systems as renewable generation grows, but organizations also need a way to measure the environmental footprint of AI itself. Garg is involved in an IEEE working group developing a standard methodology for measuring impact across electricity, energy, resources, water, and carbon.

The physical frontier is already visible in autonomous robots and drones, including systems tasked with inspecting high-risk machinery. The next phase may extend industrial AI further into the physical world, from robots and drones to AI-assisted coding that lets domain experts build new applications. Garg says these systems could change work in plants, power systems, and mining sites.

“Autonomous systems, whether they’re robots or drones, are really going to change the way that we work in plants, in power systems, on mining sites,” says Garg. “In a way, that will make these types of operations more efficient, much safer for the human beings involved and more productive.”

Current Stage: From Deployment to Redesign

The current stage is not simply a technology rollout. It is a rethinking of business processes, safeguards, and workforce roles. Industrial operators with mission-critical assets are asking how to use powerful but less predictable AI while preserving reliability and accountability. Garg’s framing is blunt: “How do we leverage these technologies while maintaining safety, while maintaining reliable operations, while still being able to deliver on the promises of the new capabilities?”

The next phase will likely be judged by how well companies turn AI capability into governed capability: better data foundations, clearer decision boundaries, sustainability measurement, and human oversight that stays central in high-consequence operations. The goal described in the discussion is a model of automation that is not only more autonomous, but safer, more efficient, and more sustainable.