In manufacturing, the stakes aren’t just high—they’re existential. It’s not a sector where you can afford to be wrong, even once.
For Christopher Nguyen, that’s exactly where industrial AI is poised to fail: when it can’t make the leap from being smart to being consistently, relentlessly right.
“What we’re facing is a crisis of lost critical expertise,” Nguyen said to a room full of manufacturing industry insiders at Stanford. “We’ve got a generation of people who know how to fix the things that break, and they’re walking out the door with all that knowledge still in their heads.”
In other words, it’s not the machines that are the problem—it’s the people. Or more accurately, the vanishing act they’re about to pull off, taking decades of embedded knowledge with them.
Nguyen illustrates via a comparison of TSMC in Taiwan versus Arizona: Same tech, same budget, vastly different outcomes. “The right-hand side has been a failure and the left-hand side has been a huge success,” he said. “And the only variable is the people involved.”
His view is guided by a rich set of experiences: a veteran technologist with a career spanning leadership roles at Google, Panasonic, and multiple AI startups, including Aitomatic, where he focuses on operationalizing human expertise through AI systems designed for industrial applications,
He’s become well-known as a vocal advocate for AI systems that don’t just process data but reason through complex, high-stakes industrial processes—especially in sectors like semiconductors, logistics, and battery production.
And Nguyen doesn’t mince words:
If you think AI is the solution to every problem, you’re not paying attention. In industries where a single missed step can mean a factory line grinding to a halt or a plant spewing toxic gas, generative AI is more liability than asset.
“Think of these LLMs as fresh PhDs,” he said. “They know a lot, but they don’t know what to do when things go sideways.”
Because that’s the thing about industrial processes: they’re not a sequence of predictable inputs and outputs. They’re full of weird little exceptions, quirks, and footnotes that only the people who’ve been solving problems for thirty years know how to handle.
Nguyen has lived this reality. When he ran AI projects at Panasonic, predictive maintenance was less about data and more about the people who could interpret it. “We actually ran into problems that AI could not solve without consulting with the one or two experts that Panasonic has in refrigeration,” he said. “That knowledge wasn’t in any manual. It was in his head.”
That’s the disconnect Nguyen is hammering on. We talk about AI as if it can be fed a thousand data points and spit out a solution. But data without context is just noise. And the people who can translate that noise into signal are aging out of the workforce. Or, worse, they’re not talking.
At Panasonic, it was refrigeration. In the semiconductor world, it’s plasma chamber fluctuations. Nguyen described a project involving fault diagnosis AI where the goal was to sit down with a soon-to-retire engineer and pull every scrap of hard-won knowledge out of his head before it disappeared forever. “I’m not saying that figuratively,” he emphasized. “I’m saying that actually.”
In the fluctuation case, the engineer made a connection no one else could have—the pressure dip traced back to a gas line installed a week prior, a tweak so small it wouldn’t have been documented anywhere. But it was enough to cause a yield issue. “That knowledge wasn’t in any manual,” Nguyen said. “It was in his head.”
That’s not an isolated example either. Nguyen points to battery production lines, where scaling predictive maintenance is less about detecting catastrophic failures and more about picking up on subtle, cumulative anomalies–things that could go unnoticed for weeks but eventually snowball into costly downtime.
And it’s why Nguyen is adamant that AI alone isn’t enough. You need the people who know how to read those signs.
“For the first time, I can get an expert to sit down with me with a machine and just start dictating,” Nguyen said, describing what he calls the “capture and apply” model. The AI listens—not just to the words but to the underlying logic, the if-this-then-that framework that turns a collection of observations into an operational guide.
That’s the principle behind Semic Kong, an AI initiative born out of the AI Alliance. The project uses modular, domain-specific agents to monitor semiconductor manufacturing processes, integrating real-world expertise into AI frameworks.
“For those of you that don’t come from the semiconductor industry, this is highly unusual,” Nguyen said. Unlike software, where open-source models are commonplace, semiconductor manufacturing is fiercely proprietary, guarded, opaque. Semic Kong aims to crack that open—not by sharing trade secrets but by operationalizing expertise that, until now, has lived in engineers’ heads.
But the question remains: How do you operationalize expertise without the expert?
Nguyen doesn’t hedge. AI systems, as they’re built today, aren’t equipped to do that. They can process data, yes. But they can’t reason through a problem like a person with a plan.
“LLMs as they are today don’t have the loop. You still need something outside of that loop to say, okay, given that the world has changed, what do I do about it?”
That’s the pivot. AI can make predictions, but it can’t plan. And that gap is where Nguyen is focusing his efforts—on building frameworks that don’t just spit out answers but think through problems as new data emerges.
“It’s not about brute force,” he said, “it’s about intelligent persistence.”
And that’s what industrial AI needs to be: not just a prediction engine but a reasoning engine, a system that doesn’t just know how to identify a problem but knows what to do next.
In the semiconductor fabs Nguyen describes, the AI needs to understand not just that pressure has dropped but that it might be linked to a specific change in the gas line last Tuesday. And that connection isn’t data. It’s experience.
The same applies to battery production, logistics, and manufacturing across the board.
Nguyen calls this a “revolution,” but he’s not talking about a revolution in hardware or algorithms. It’s a revolution in mindset—a shift from thinking of AI as a source of answers to thinking of it as a structured, operationalized version of what the best engineers already know.
“The opportunity is in that domain expertise. We always knew it was valuable, but the reason we haven’t used it is because it’s too difficult to encode.”
Difficult. Not impossible.
Nguyen’s warning is clear: companies that fail to operationalize that expertise will lose it. And when it’s gone, it’s gone.
“Lead the industrial AI revolution—don’t wait for someone else to do it,” he said. “The opportunity is here, but it’s only as real as your willingness to act.”