For years, the Industrial IoT fixated on the singular idea of connecting everything. Investments centered on connecting machines via sensors and aggregating data to give operators greater visibility into operations. And for a while, that was enough.

Visibility, it turns out, is table stakes. That phase is now giving way to a more consequential competitive divide focusing on what happens after data is captured. Call it digital intelligence.

Now, the challenge is to turn raw signals into real-time decisions. Technologies such as intelligent edge analytics, anomaly detection and live digital twins help organizations understand, predict and respond as events unfold. Capturing the data has become relatively straightforward. Turning that data into operational decisions is the bigger challenge.

Barriers to Scaling Industrial AI Across Multiple Sites

Despite the momentum, enterprise adoption remains uneven. Subramanian points out that the gaps are structural, starting with what he describes as uneven data foundations. One plant’s baseline is another plant’s blind spot. Inconsistencies in equipment, fragmented control systems and uneven instrumentation make standardization elusive. “Plants differ widely in equipment age, control systems and data maturity—what’s instrumented at one site often doesn’t exist at another,” he said. 

The second barrier is the lack of a reusable deployment model. Too many organizations still treat each rollout as a one-off. In contrast, leading manufacturers standardize early. “They baseline asset archetypes early, identify which equipment families behave similarly across sites and templatize the sensing, modeling and response logic into reusable frameworks,” he said.

“That turns a 12-month plant deployment into a six-week configuration exercise. We see this work consistently across global consumer goods process manufacturers, where a templatized smart manufacturing approach has saved thousands of person-hours per program and delivered ROI well within the first year of each rollout—because the accelerator templates eliminated repeated reinvention across sites.”

The third barrier is organizational. Data ownership, infrastructure management and operational accountability are often fragmented across silos. Scaling AI, Subramanian argues, requires “shared ownership of results, not just shared access to a platform.”

How to Close the Gap Between Data, Decisions and Action

If the goal is to reduce the time between detecting a problem and resolving it, it is fair to ask whether that trajectory ultimately leads to autonomous operations.

Subramanian believes it does, but not in the way that many vendors suggest. Meaningful autonomy doesn’t come off the shelf and rarely does it emerge from a single software. Instead, it emerges from purpose-built architectures designed around a facility's existing operational systems, data flows and decision-making processes. In other words, autonomy is less a technology purchase than an engineering challenge—one that depends on how effectively data, controls and business systems are connected across the industrial stack.

READ MORE: How AI Transforms Fragmented Data into Actionable Engineering Intelligence

“Autonomous response requires AI that can talk across the full operational stack—from warehouse execution systems (WES) and warehouse control systems (WCS), through manufacturing execution systems (MES), up to ERP and enterprise planning layers,” said Subramanian. “No single product does that out of the box. It takes agentic system design: AI agents with defined roles, decision boundaries and handoffs, built to orchestrate actions across those layers in sequence and in real time.”

Agentic AI extends beyond assistants by pursuing goals, making decisions and executing tasks across systems. It coordinates actions with minimal human input. Over time, this model evolves into networks of specialized agents to execute tasks based on human intent.

How Agentic AI is Orchestrating Industrial Workflows

This shift toward orchestration is now emerging in industrial systems through agentic design.

Siemens has piloted Eigen with more than 100 companies in 19 countries. In the U.S., Prism Systems used it to create, modify and import SCL code in seconds. “The challenge has been bringing that capability into real industrial workflows,” said John Elias, President at Prism Systems. “Siemens’ latest tools help close that gap, allowing us to apply AI in a way that truly supports engineering and automation.”  

While the breakthrough is notable, its transformative potential lies in the architecture it enables—the ability to orchestrate decisions across interconnected systems. As routine tasks become automated, engineers can focus on more complex system-level challenges. Siemens reports the agent executes AI-powered workflows two to five times faster than manual approaches, improves solution quality by up to 80%, and raises engineering efficiency by 50%.

What it Takes to Move Industrial AI from Pilot to Scale

Gartner reported that by the end of 2025, at least 50% of generative AI projects were abandoned after proof of concept due to poor data quality, inadequate risk controls, escalating costs or unclear business value. The finding suggests that the challenge facing many organizations is not AI technology itself, but the ability to operationalize it at scale.

That raises a broader question for industrial enterprises: What separates AI initiatives that remain trapped in pilot mode from those that evolve into trusted, production-grade systems capable of supporting operational decisions?

Subramanian said a mature AI operating model is clear about four things: which decisions AI supports, where humans stay in the loop, how models are monitored and retrained, and who is accountable for outcomes in business terms. “What’s almost always missing early is that last one,” he said. “When AI is treated as an IT initiative, it tends to plateau in pilot mode. When it’s owned by the leaders responsible for yield, uptime or cost—and success is measured in those terms—it scales.”

READ MORE: Beyond the Sticker Shock: The Real Reason Companies Hesitate to Adopt Robotics

If this diagnosis holds, the implication is clear: Autonomy is not a standalone technology upgrade, but an outcome of how effectively industrial systems are integrated and governed. Fragmented systems, unclear decision ownership and siloed accountability continue to slow adoption. In this context, execution speed (how quickly organizations can move from sensing a condition to taking action) becomes a defining competitive differentiator.

In Subramanian’s view, many companies undermine AI initiatives before they reach scale. They are chasing the destination before building the road to get there. The journey has three phases: The first phase focuses on targeted productivity gains, proving value in specific use cases. The next phase—industrialization—integrates AI across the stack, standardizes data and treats model performance as an operational responsibility. The third phase introduces AI agents to orchestrate decisions across the enterprise.

The common misstep is skipping the industrialization phase in favor of premature autonomy. The industrialization phase is where measurable value is realized at scale, pointed out Subramanian. This architecture-first approach is reflected in large-scale deployments such as Cognizant’s smart manufacturing work with a global industrial tools manufacturer, where more than 100 plants and roughly 1,000 machines were connected through a cloud platform without disrupting operations. The program delivered approximately $200 million in value over three years by shifting IIoT from monitoring to enterprise-wide operational decision-making.

Trust also remains critical. “Operators don’t resist AI because they don’t understand the technology; they resist when the system makes decisions they can’t see into and can’t override,” said Subramanian. “Designing explainability and human override into the architecture from the very first pilot—not bolting it on later—is what makes autonomy trustworthy rather than threatening.”

What it Boils Down to for Design Engineers

For all the attention being paid to AI, digital twins and autonomous operations, the real competitive divide is increasingly clear. The organizations pulling ahead are those building the architectures, workflows and accountability structures that allow insight to become action.

For design engineers, that shift begins much earlier—in decisions about sensors, controls, connectivity and interoperability that determine not only whether machines can participate in a larger operational ecosystem, but how effectively they do so. As Subramanian observed, “Trust in AI follows demonstrated value and demonstrated value requires someone accountable for the outcome, not just the deployment.”