If you’ve invested in AI, machine learning, and digital twins only to see disappointing returns, Episode 14 of Practitioners Unplugged diagnoses the problem. Craig Henry, Global Account Director for Amazon at Murrelektronik and author of Super Connected: The Future of Industrial Nervous System, cuts through the AI hype with a sobering truth: your factory’s expanding “brain” is useless without a properly designed nervous system to connect it to reality.

With over 30 years in advanced manufacturing and inter-logistic systems—including major roles at Amazon, Siemens, and Danaher—Craig brings a practical, infrastructure-focused perspective to Industry 4.0 and 5.0. His book and this conversation challenge the current obsession with AI capabilities by asking a more fundamental question: can your AI actually feel what’s happening on your factory floor?

As Craig puts it: “We’re operating a brain that is suspended in space and isn’t connected to a body. Without the nervous system, we are crippled.”

Here are the five key insights from our conversation with Craig:

While manufacturers pour resources into AI analytics platforms (the “brain”) and field devices (the “extremities”), Craig identifies a critical gap: the connectivity layer that bridges them. This “forgotten middle” represents the difference between AI that transforms operations and AI that generates impressive demos with no business impact.

“As I’ve talked with Amazon and others in the field, I’m seeing that when you get the connectivity right, it opens up all of the promises of Industry 4.0 and Industry 5.0 and the digital twin. But without it we are crippled.”

The problem is pervasive: manufacturers operate with 30-year-old control systems that create “black box effects.” A CEO cannot drill down through cybersecurity, edge infrastructure, and proprietary PLCs to see a single sensor reading that might explain why a critical customer order is delayed. The data exists but remains invisible.

Craig’s prescription: invest in the forgotten middle before investing in AI. Move from analog signals to digital communication networks. Implement open architectures like OPC UA that expose data across the organization. This infrastructure investment delivers the fastest ROI because it enables you to see and feel what’s happening across your operations for the first time.

The parallel to Amazon’s approach is instructive: Jeff Bezos mandated 20 years ago that every system would have an open API with exposed data. This infrastructure-first philosophy enabled Amazon’s rapid innovation and operational excellence—not because they had better AI, but because their AI could actually see the operation.

Craig addresses the objection every plant engineer faces: why spend $20,000 installing a $200 sensor when budget pressures demand “value engineering”? His answer reframes the investment from cost to insurance against catastrophic failure.

“I would venture to say having that data could mean you’ve saved a million dollar problem down the line from happening. And the $20,000 would be small compared to what could be just operating blindly and hoping for the best.”

The critical distinction: don’t install analog signals—install communication networks. Run digital, self-checking, deterministic, error-checking protocols through that expensive conduit run. This enables condition monitoring that moves unplanned downtime to planned maintenance, a transformation that pays for sensor investments many times over.

Craig cites the harsh reality of “value engineering” in capital projects: data infrastructure gets cut first, then costs 30-50% more to retrofit later when problems emerge. Leading organizations take the opposite approach—they get data first, even before knowing exactly how it will be used, because they understand the infrastructure enables future capabilities.

The lesson extends to OEMs and system integrators: companies that deliver equipment with comprehensive digital twins and exposed data differentiate themselves from competitors still selling hardware. The conversation shifts from price to ROI when you can demonstrate how your system will improve throughput by specific percentages.

Craig challenges both the “lights out factory” dream and the passive “human in the loop” compromise with a more sophisticated paradigm: human in command. This distinction becomes critical as automation advances and skills fade threatens operational resilience.

“It’s not human in the loop, but human in command. The system should be able to push up to him: I’m seeing this. Here are three options. Please choose one and approve it.”

The analogy to chess masters illustrates the point: the highest performance comes from humans with computer assistance, not computers alone. The master can recognize patterns, apply judgment, and execute strategies that algorithms miss. Similarly, factory operators should manage 20 stations with AI flagging exceptions and recommending actions rather than pushing buttons on individual machines.

This approach addresses multiple converging crises: the global labor shortage (500,000 positions unfilled in U.S. fulfillment alone), the competency crisis (87% of companies cite skill gaps as their biggest constraint), and the skills fade problem (automation that works well until catastrophic failure when humans have stopped understanding the underlying process).

Craig’s warning about autonomous driving accidents applies equally to manufacturing: when humans assume automation has everything under control, disasters happen. The solution isn’t eliminating automation but designing systems where humans maintain operational understanding and decision authority even as AI handles routine optimization.

Craig takes a strong stance on the standards debate, arguing that proprietary protocols and closed systems amount to holding customers hostage. For manufacturers evaluating equipment purchases or system upgrades, this represents a critical strategic decision.

“If I decide, hey, I’m going to force my customers who buy my equipment to use my protocols and my networks to the exclusion of everybody else, what they’re really doing is holding their customer for ransom. And that’s not cool. It’s not okay.”

His recommendation is clear: use OPC UA, which bridges devices to cloud, includes built-in encryption, and democratizes access across vendors. Avoid fieldbus systems controlled by single vendors. Demand that OEMs expose data through standard interfaces even if internal IP remains protected.

The business case for open standards extends beyond technical elegance. Standards enable faster deployment, broader talent pools for implementation and support, and protection against vendor dependency. When Amazon and other large end-users mandate specific protocols, they’re protecting their ability to innovate independent of equipment vendor roadmaps.

For OEMs, Craig’s advice flips the conventional wisdom: don’t view open standards as giving away competitive advantage. Instead, differentiate through consulting on process improvement and ROI rather than locking in through proprietary interfaces. Companies that help customers achieve measurable performance gains win business regardless of protocol choices.

Craig delivers sobering statistics: cybersecurity attacks now generate more revenue than illegal drug trafficking, with loss of use (ransomware) as the primary attack vector. For manufacturers, this isn’t just an IT problem—it’s an operational resilience issue that requires scenario planning and regular drills.

“We have to be able to have plan A, plan B, plan C, and to your point that a human is there skilled and experienced enough to be able to drive that and to make those decisions.”

The challenge intensifies with automation dependence. Craig’s example of his wife’s county clerk office illustrates the problem: when systems went down, resistance to manual processes was so strong that basic operations nearly stopped despite manual procedures being available. In manufacturing, returning to manual operation may be impossible when robots perform tasks no human workforce could replicate.

The solution requires acknowledging that attacks will happen and preparing accordingly: secure data in three places, maintain operational knowledge that doesn’t depend on systems, and ensure humans understand underlying processes enough to make decisions during disruptions. This represents the flip side of the human-in-command principle—humans must maintain enough competency to take control when automation fails or becomes compromised.

Craig’s perspective on large language models adds another dimension: 80% of people using LLMs can’t recall the content they produced. If we build similar dependency in manufacturing operations, how do humans respond when AI systems become corrupted? The answer requires intentional design to maintain human competency even as automation advances.

Craig Henry’s infrastructure-focused approach provides the reality check that digital transformation initiatives need. While the industry obsesses over AI capabilities—with projects like Stargate committing $400 billion to expand the “brain”—the nervous system that connects that brain to physical operations lags dangerously behind.

The key lessons for practitioners: invest in the forgotten middle connectivity layer before chasing AI solutions, move from analog signals to digital communication networks throughout your operations, design for humans in command rather than lights-out automation, demand open standards that prevent vendor lock-in, and prepare for disruption through scenario planning and competency maintenance.

As Craig warns, “We have an uninformed brain floating in space, thinking it understands things.” The physical world of manufacturing requires more than thinking—it requires feeling, sensing, and responding through a properly designed nervous system.

The path forward isn’t abandoning AI but building the infrastructure foundation that makes AI useful. Organizations that invest in super-connected operations—with digital networks exposing real-time data, open standards enabling scalability, and humans maintaining command authority—position themselves to leverage whatever technologies emerge next.

The revolution isn’t in the AI capabilities themselves but in the infrastructure that lets those capabilities touch reality. Build the nervous system first, and the brain’s potential becomes achievable. Skip this step, and your AI investments remain impressive demonstrations disconnected from operational impact.

The question isn’t whether to invest in digital transformation but where to invest first. Craig’s answer is clear: start with the body your AI brain needs to function.

To submit a request for a new episode topic from Practitioners Unplugged,
visit our contact page

Thanks for reading. Don’t forget to subscribe to our weekly newsletter to get every new episode, blog article, and content offer sent directly to your inbox.

Explore more about Schneider Electric & AVEVA