For years, we have enhanced automation systems much like adding GPS or cameras to a car. These upgrades deliver incremental value, but just as adding navigation and a camera doesn’t turn a classic car into a self-driving vehicle, autonomous operations need a stronger foundation. That foundation is contextualized, asset-centric information organized within a clear plant hierarchy, which makes scalable, integrated, AI-driven autonomy possible. Without that, automation systems cannot overcome the limitations of the underlying architecture.
That’s roughly where industrial automation stands today. AI can already help engineers work faster, operators make better-informed decisions, and plants optimize specific processes. The longer-term destination is autonomous operations. The question is: how do we get from one to the other without abandoning the systems, knowledge, and investments that got us here?
For most plants, the answer won’t be a leap. It will be a series of deliberate, incremental steps that build on existing systems, knowledge, and investments while creating a foundation for greater autonomy over time.

Prepare for industrial AI by digitizing plant knowledge
For many brownfield facilities, the first AI opportunity is surprisingly unglamorous: digitizing what the plant already knows. Process knowledge may be scattered across drawings, handwritten redlines, manuals, operator journals, and employee experience. AI-assisted tools can help turn that material into structured digital information, with human review helping interpret plant-specific information consistently.
This cautious, human-in-the-loop approach reflects a broader industry trend. Generative AI is already being explored for manufacturing applications such as root-cause analysis, troubleshooting, and operating guidance, but human operators must remain actively involved in its use.
Once contextualized, that knowledge becomes much more useful. An operator investigating an alarm could identify affected equipment, retrieve relevant documentation, explore possible root causes, and consider recommended actions. The judgment and action remain with the operator.
That depends on trusted, validated data. AI models need to operate within known boundaries, drawing from reliable plant knowledge while experienced people validate and strengthen their outputs.
AI can also preserve expertise. When veteran engineers and operators teach models how their plant works and reinforce preferred practices, they’re capturing practical knowledge that might otherwise leave when they retire.
How software-defined automation creates a foundation for industrial AI
Turning existing plant knowledge into trusted digital information is an important first step. The next challenge is making that information accessible and usable across the automation lifecycle to support engineering, operations, maintenance, and system management.
EcoStruxure™ Foxboro Software-Defined Automation (SDA) extends this human-in-the-loop approach beyond knowledge capture and into automation execution. Built on EcoStruxure Automation Expert, its open, software-defined architecture helps connect engineering and operational information across the lifecycle, providing the context AI needs to deliver meaningful results.
In traditional automation environments, critical knowledge is often fragmented across engineering tools, documentation, control applications, and individual experts. Software-defined automation helps break down those silos, enabling AI to work with a more complete view of plant information while keeping engineers and operators in control of decisions. Some of those capabilities are practical today.
For example, AI can use digital piping and instrumentation diagrams, process flow diagrams, and cause-and-effect information to help create a preliminary control application. Engineers can then review, validate, and refine the results using the customer’s preferred practices for configuring pumps, reactors, heat exchangers, and other assets.
AI can also examine existing application logic and identify opportunities to improve performance and maintainability. A function block, for example, may have become unnecessarily complex or take longer to execute than necessary. Breaking that logic into smaller, more modular functions could improve both execution efficiency and long-term maintainability.
The objective is not to replace engineers. By reducing time spent transcribing information, searching documentation, and recreating existing work, AI can help engineers focus more on design decisions, operational improvements, and the application of their expertise.
From AI recommendations to process optimization
Historical data can tell us what happened, but not always why. Building a useful brownfield model may also require operator action journals or other records showing which human interventions produced those outcomes.
Once teams have a sufficiently useful model, they can take it offline and apply reinforcement learning. By defining the desired outcome—perhaps throughput, quality, energy performance, or reduced variability—along with operating constraints, the model can explore different strategies without experimenting on the live plant.
In Schneider Electric’s Foxborough, Massachusetts lab, engineers used reinforcement learning on a simple closed-loop tank process and demonstrated tighter variability than a conventionally tuned PID loop. While this is a laboratory example, not a case for replacing PID control wholesale, it shows the potential to operate closer to an optimum while independent safety systems remain in place as essential guardrails.
Building more autonomous and resilient operations
Process optimization is only part of the autonomy journey. To move closer to autonomous operations, the automation system itself must become more resilient and increasingly capable of managing routine disruptions.
High availability disruption avoidance, for example, uses orchestration to respond to a controller failure by moving control workloads to available hardware and restoring redundancy with less manual intervention. That introduces the idea of a system that can help maintain itself, another step toward the broader zero-touch automation vision.
However, “toward” is the key word. The industry isn’t suddenly handing the keys to AI. We’re building capability, testing it, learning where it adds value, and becoming more comfortable with what it can reliably do.
A practical approach to industrial AI
For customers interested in exploring AI, a useful first step is often a digital transformation engagement. Rather than beginning with the technology, start by understanding where it can create measurable value:
- Identify the obstacle or opportunity. What operational problem are you trying to solve, or where could greater intelligence improve performance?
- Understand the information you already have. Review available process data, documentation, operator knowledge, and data quality to determine what can support the use case.
- Match the use case to the architecture. Determine what the current automation environment can support and identify a focused application where AI can add value without requiring an immediate leap to full autonomy.
- Establish the expected return. Understand the potential operational or financial value before committing to new hardware, software, or broader implementation.
From there, customers can start small, gain experience, measure the results, and carry successful investments forward as their automation architecture evolves.
Start with value. Build trusted information. Keep people in the loop. Add capability deliberately. Then let evidence—not enthusiasm—determine when it’s time to take another hand off the wheel.
The move to AI-enabled operations
Building a foundation for industrial AI starts with connecting engineering and operational data across the automation lifecycle.
Learn how EcoStruxure Foxboro SDA helps create the contextualized, software-defined environment needed to support AI-driven engineering, optimization, and autonomous operations.
Add a comment