Actors and musicians have agents that handle all the negotiations with film studios and record companies. The performers don’t have to worry about any of that nonsense. The agent does the work for them.
Agentic AI, or AI that enjoys autonomy based on data analysis and learned behavior, might do the same for your factory tools.
Imagine a lathe with an agent that sets its spindle speed and how often a robot will feed in new workpieces. Rather than simple stop-or-go decisions, the AI could make nuanced calls like slowing down when temperatures get too hot, conducting vibration analyses to predict faults and working with the maintenance department’s agentic AI to schedule preventative service.
It’s not right around the corner, but this is not fantasy talk either. If manufacturers adopt the technology en masse, the impact on who does what in the industry could be profound.
I spoke with Aaron Merkin, CTO at Fluke Reliability, at the Xcelerate 2025 conference last week about agentic AI and what it might mean for the plant floor. The interview is edited for clarity and length.
Aaron Merkin: Traditional process control systems have been the orchestration style of “Let me define the process flow that I want.” What’ll be interesting to see is if agentic AI creates the ability for individual nodes in that process control flow to start acting more autonomously and switch to a choreographed model. [That] allows you to have much more dynamic operations.
In the case of a factory, a machine may potentially respond if a fault condition arises. If a machine anticipates a conveyor belt’s supposed to be feeding at 80 feet per minute, and it slows down to 40, what does the equipment that’s next to the conveyor do?
In a traditional process control, you have an operator who recognizes trigger conditions and [slows down a machine].
In an agentic AI model, rather than a trigger telling a control function to [slow down the machine], you actually have the machinery recognizing that the input speed has slowed down.
Dennis Scimeca: If each station in a production line can control itself and read what’s happening before and after it on the line and make its own decisions, doesn’t that take the human entirely out of the equation?
Let’s say you have someone monitoring the entire process on the floor and agentic AI decides to slow down a machine. Maybe that is or isn’t the right decision, but it’s already been made. Needing a human to validate every tiny correction defeats the purpose of deploying AI for speedier decision-making. So where does the human come into the loop?
AM: In process control today, you have human beings who, instead of being the ones who respond and make the changes, are on the loop, monitoring. And [even with the rise of agentic AI] I think that you’ll still have that need to set boundaries for the range of decisions that the autonomous nodes can make.
That’s where human expertise will be relied on for a longer period of time… tuning the agentic AI behaviors based on the human’s understanding of what can and can’t go wrong.
DS: Manufacturing leaders are constantly concerned about the inability to replace retiring, veteran workers. Let’s say you have a departing operator tweaking the agentic AI to tune it just right. Couldn’t that be looked at as knowledge transfer? And doesn’t that trigger the concerns about AI replacing people?
Some parameters are still going to change…but at a certain point, the agentic AI might reach a critical mass of knowledge where you can rearrange the floor, make changes to processes, and the AI has learned enough that it can adapt without needing those tweaks. The AI could self-correct, based on its experience, to fresh changes.
AM: I think it’s unfortunate what’s going to happen to low- to medium-skilled jobs. We can talk about vibration analysis. The [basic data collection and trend analysis] jobs are going to go away because AI can do those.
But [for more complicated analysis] you’re still going to need senior experts evaluating the decisions made by AI. … You can’t program it and set an outcome and know what you’re going to get. You adjust the probabilities as best you can so that the results stay within the threshold of what’s acceptable.
I think there’s two competing things. You’re going to have a hollowing out of the need for lower-skilled operators but a demand for much higher-skilled expertise to oversee the AI.
The longer-term concern I have…is that the [manufacturing] industry is going to try to reap the benefit of [replacing low-skilled operators]. And it’s not going to be in the next five or 10 years that we have an expertise gap. It’s going to be the generation after that, because we won’t have trained enough people to become the experts who oversee the systems.
DS: In a world where low-skilled jobs are replaced by either cobots on the floor, for basic manufacturing operations, or AI for basic analysis functions, what will starting positions at manufacturing companies look like?
AM: If you think practically about manufacturing, you have heavy sunk cost in capital equipment, and the effort to retrofit robotic automation into a factory, the capital cost there, is going to be so great that factories will not do that. They’ll focus on the soft skills that can be replaced with software, and they’ll continue to have human [machine operators].
DS: What about cobots? They’re small enough to occupy the same space as a human being, can handle tasks like simple torquing and welding, can be mounted on casters to move them around the plant as needed…
AM: I think that’s fair and this is really going to be the question. The CEO of Anthropic published a great essay called Machines of Loving Grace. The first part of it is a very utopian [vision of] all the advancements that AI will unlock… But then you move further afield, and what happens with the economic disruption when human beings are no longer needed to do work?
I’m less worried about [cobots]. I’m worried about the hollowing out of white-collar jobs. We’ve, for 30 years, put so much emphasis on going to college that people are going to want white collar jobs. They need to pick the thing that’s harder to replace, which is skilled trades.
[Manufacturers need to] redirect people into that and make that both attractive to people but also financially rewarding enough that they’re able to make a living and take care of their families. I think that’s probably the bigger cultural shift in industry. They have an opportunity to be in front of [the problem] and try to find ways to bring people, bring Gen Z into the trades, versus trying to bring them into back office, white collar positions.
