AI Won't Evolve On It’s Own
How to think about a technology that doesn't exist yet
Every morning, I walk around the lake and listen to podcasts. It's where most of my thinking happens, just movement and ideas.
Recently, I was listening to Robert Tercek, a media futurist, and he told an amazing story. In the year 2000, he left a full-time job to join a startup and work on a pocket-format video streaming. Everyone around him said it was ridiculous. Back in 2000, the data networks were too slow, and screens were too small; basically, the infrastructure didn't exist. So all the sceptics were right, back then in 2000.
Now we are in 2026, living in a world where billions of people carry a small cinema in their pocket, watching video content, while waiting for a coffee. Not that ridiculous anymore (not normal either, but here we are).
Since Tercek wasn't wrong about the destination, I couldn't stop thinking that this is exactly where we are with the Industrial AI on the factory shop floor. The sceptics again are right, about today. AI makes mistakes, the data is messy, integration takes months, and ROI doesn’t always match the reality on the shop floor. Anyone who tells you Industrial AI is already solved hasn't spent enough time in a real plant.
But being right about today is not the same as being right about direction. The question I ask is not whether Industrial AI is perfect now, but where it is going, and as someone building technology solutions for manufacturing, how can I think about it with clarity?
Tercek, in his podcast, brought up Marshall McLuhan, a Canadian philosopher who spent his career studying how new technologies reshape human behaviour and society. He identified four things - a tetrad - that every new technology does, without exception.
It amplifies something
It makes something obsolete
It retrieves something we thought was lost
And when pushed to its limit, it reverses its own effect
I looked into this, and I tried to find symmetry in Industrial AI. What it reveals is more interesting than most AI commentary I read, and more useful for anyone trying to make real decisions about where to place their bets in manufacturing.
Let me walk you through all four.
What Industrial AI amplifies
Every technology extends something human. The microscope extended sight. The telephone extended voice. The car extended legs.
Industrial AI extends perception.
This is the most important thing to understand about what this technology actually does, and most people miss it because they stop at productivity. "AI makes us more productive" is true but shallow. It's like saying the microscope made scientists faster. Technically correct. But it misses the point entirely.
What AI actually does on the factory shop floor is extend the human senses beyond what any individual — or any team — could achieve alone.
Consider what that means in practice.
A human analyst can monitor a set of process parameters and spot patterns within their experience and attention span. An AI system monitors thousands of parameters simultaneously, continuously, without fatigue, and finds correlations no human could hold in mind at once. It doesn't just see more. It sees relationships that were always there but invisible.
This is pattern recognition beyond human bandwidth. At Advanced Solutions, this is the foundation of everything we build. Not faster reports. Not better dashboards. A fundamentally different level of visibility into what is actually happening inside a process.
The second extension is memory. Human expertise is fragile because it retires, resigns, or moves to another factory for a few cents more per hour. Decades of compound knowledge, gone. AI extends institutional memory indefinitely and makes it organised and available. The plant doesn't forget when the expert leaves.
The third is attentiveness. Of all human capacities, attention is the one that degrades fastest under industrial conditions. Shift work, repetitive monitoring, alarm fatigue. These are not human failures; they are human limits. AI extends attentiveness without fatigue, and that’s why quality control, anomaly detection, and predictive maintenance are all fundamentally problems of sustained attention that AI is structurally better equipped to handle.
The fourth, which is our focus at Advanced Solutions, is the velocity of output. The time between an incident occurring and a decision being made. In most manufacturing environments today, that gap is measured in hours, sometimes days. Shifts change, engineers are unavailable, reports are written, and meetings are called. By the time a decision is made, the process has drifted further, the quality has escaped, and the cost has compounded.
We compress that gap from weeks to seconds. It was possible because we removed the steps that existed only due to human limitations in the first place.
Industrial AI enhances perception, memory, attentiveness, and the speed of applied intelligence. These are the four extensions that really matter, and productivity is the result that follows.
What Industrial AI is making obsolete (and what it is not)
Obsolete doesn't mean dead. It means no longer in charge.
The car didn't kill walking. It displaced walking from the centre of how we move through the world. The same logic applies here. When I say Industrial AI is making things obsolete, I don't mean people disappear. I mean, certain roles, structures, and systems lose their authority, their position as the essential link in the chain.
Understanding this distinction is important. Because most of the fear I observe in manufacturing is aimed at the wrong target.
When AI entered the film industry, screenwriters organised protests. They argued loudly and publicly that their craft was under threat. But the creative layer, the imagination, the narrative judgement, the ability to tell a story that moves people, is exactly where writers excel, and AI doesn't. What AI actually threatens in film is the cost of execution. The physical sets, the location logistics, and the enormous production apparatus, which are required to turn a creative vision into something on screen, can suddenly be replaced by a single creator from behind his desk. Not perfectly yet, but the direction is clear.
Manufacturing has the same structure.
Process engineers, quality managers, or technical support specialists are defending their authority visibly and sometimes aggressively. I observe this first-hand. And I understand it. When you have built your value on being the indispensable expert, the arrival of a system that can replicate some of that expertise feels like a direct threat.
But the engineer is not what is actually being obsolesced here. It’s the scaffolding built around human limitations. The quality could not have been monitored continuously, so the spot quality checks were invented. The shift handover reports were written by hand because knowledge didn't transfer automatically. The escalation chains existed because decisions required the right person to be available. The months of training investment that walk out the door when an operator moves to another factory for a few cents more per hour. The endless reinvention of solutions for problems that have already been solved three times in the last six months. In the context of Industrial AI, I see the engineer not being the production crew, but more like the screenwriter.
The ability to read a process, to understand why a material behaves the way it does under specific conditions, to make judgements that require years of accumulated pattern recognition — that expertise is irreplaceable. That is exactly what AI needs to work with. Not instead of.
AI makes that existing scaffolding obsolete, because that expertise and knowledge were hard to move, hard to scale, and hard to keep.
So the question for anyone in a knowledge-intensive manufacturing role isn't how do I defend my position? It's how do I separate my expertise from the scaffolding — and focus entirely on the former?
The engineers who decide to make the transition early will build systems for shift handovers instead of writing them. They will design the escalation logic instead of being the escalation. They will contribute to a knowledge library that captures what you know before it retires. They will move from firefighting to architecture.
Those engineers will not be displaced; they will be the ones who build what comes next.
There is one more thing worth saying. The data access will be democratised. Every plant, every supplier, every operator will eventually have access to the same knowledge, the same models, the same tools. That equalisation is coming, and it is unstoppable.
The competitive advantage will not come from having access to better data. It will come from having designed systems for speed before everyone else did. The organisations that built the architecture early and replaced the scaffolding deliberately will have a lead that data access alone cannot close.
The window to build that advantage is open now. It won't stay open forever.
What Industrial AI is bringing back
This is the law most people don't expect.
Every new technology reaches back. It brings something back that a previous technology buried. And what Industrial AI is bringing back is something manufacturing lost a long time ago — so gradually that most people don't even notice it's missing.
Go back before the modern factory, before Henry Ford invented a production system of repeatable steps and before the standard operating procedures became the authority. There was the craftsman.
He didn't consult a document to know when the product was right. He knew. Decades of experience, translated into instinct. That knowledge lived in his hands, his eyes, and the way he paused before making a decision.
Then, scientific management arrived. Know-how was a problem because it couldn't be standardised or transferred. So it was replaced with written procedures, defined tolerances, and repeatable steps. It worked, and it scaled, but it lost something important in the process.
Industrial AI brings that instinct back. Not romantically, but technically. Machine learning finds patterns that no one could write down. It builds judgment from thousands of observations. It detects that something is about to go wrong before any alarm is triggered. This is why pattern recognition is my development priority for 2026, as it gives manufacturing back something it traded away for scale.
But there is more.
Think about how knowledge used to “travel” between people. It was not through manuals, it was through observation. An experienced person worked, and a less experienced person observed, asked questions, and gradually understood. Today, advanced analytics systems do something similar: they watch a process, spot deviations, and suggest corrections. The system learns by observation, just like people used to.
Think about the operator who used to know the whole process, not just one task, but the entire workflow. Mass production created narrow specialisation, but Industrial AI will bring it back. Our recent development is built on this idea: one skilled engineer, with a set of agentic AI handling the micro-tasks, can do what previously required an entire team of experts.
Think about the knowledge that retires with experienced engineers leaving every year. They take thirty years of hard-won understanding with them. Most of it was never written down because it was too complex, too contextual, too human. Solutions like the knowledge libraries are designed to capture exactly this, before it walks out the door. The reaction from manufacturing leaders is always the same: relief. Because they know what it costs when it doesn't get captured.
And think about experimentation. For decades, R&D happened away from the factory floor, in separate buildings, by separate teams, within a separate reality. AI brings it back to where it belongs. Continuous improvement, tested in live production, informed by real data, iterated in real time.
What Industrial AI retrieves is simple: the connection between knowledge and the moment it is needed. That connection was broken in the name of scale. AI restores it, at scale.
When AI becomes its own worst enemy
Every technology, pushed far enough, flips.
The highway was built to speed up travel. Add too many cars, and you get a traffic jam. The solution becomes the problem. This is not a malfunction, but the natural logic of every technology taken to its limit.
Industrial AI is no different, and very few are talking about it.
More optimisation, more fragility.
AI makes processes faster and leaner. It removes waste, tightens tolerances, and reduces variation. This is exactly what manufacturing needs until the system becomes so optimised that it can no longer absorb anything unexpected. A supplier delay. A material variation. An equipment failure outside the predicted pattern. The more efficient the system, the more a single disruption costs. Extreme efficiency produces extreme fragility.
More prediction, less judgment.
AI supports human decisions. Operators and engineers make better calls because the system gives them better information. But push this far enough and something quietly breaks. People stop exercising judgment because the system always has an answer. Over time, the skill erodes. And when the system is wrong (which it will be), the humans who were supposed to catch the mistake no longer have the instinct to do so. You end up with worse decisions than before AI arrived.
More data, less understanding.
AI makes the factory floor more visible. Every parameter measured, every anomaly flagged, every trend tracked. Until there is so much signal that no one can interpret it meaningfully. One of our customers was receiving 2000 alert emails per day from his monitoring system. Two thousand. The system was technically working perfectly. But it had made itself useless. When everything is urgent, nothing is. The reversal: more data destroyed the trust that makes data valuable. Applications which can solve it do not need more information, but better information. The right alert, at the right moment, with enough context to act on it immediately. Cutting through the noise is not a feature. It is the entire design philosophy.
More automation, new bottlenecks.
AI removes dependency on manual work across most of the process. Until the few tasks that remain and the ones that still require human intervention become the new constraint. Tool change is a good example. One of our customers told us directly: even with the best AI system in place, tool change is still his bottleneck. You can automate everything around a problem and still be held hostage by the one thing you didn't solve.
But the reversal is not a reason to stop. It is a reason to design differently. The companies that understand where the flip happens, before it happens, will build systems that stay on the right side of the line. Precise enough to be trusted. Flexible enough to absorb disruption. Fast enough to matter. Human enough to remain in control. That is the design challenge. And it is harder than building the Industrial AI itself.
Where this is going
I am not a fortune teller, but I have spent enough time inside manufacturing, close enough to the technology, to make some directional bets. These are not wishes. They are conclusions from everything the four laws point toward.
One. The speed between incident and decision becomes the main competitive differentiator.
Automation can handle the repeatable, and manufacturing has been doing robotics, process control, and scheduled maintenance for decades. The unsolved problem has always been what happens when something unexpected occurs. A process drifts. Quality declines. Something breaks the pattern. At that moment, humans must step in, and the time between incident and decision is where costs compound, quality escapes, and reputations are damaged. The companies that compress that gap will gain the advantage that matters.
Two. One engineer will replace a department.
Engineers won’t disappear, but the right engineer — fluent in models, prompts, knowledge hubs and data — will be able to build an entire technical organisation using agentic AI platforms. Each agent will be able to handle a micro-task, such as shift handovers, process drift detection, reporting, and escalation logic. The competitive advantage won't be the size of your engineering team. It will be the capability of the one person who knows how to build one. What they will need is the right Industrial AI platform.
Three. Knowledge becomes the most valuable asset in manufacturing (and most companies have already lost half of it).
The retirement wave has been happening for twenty years. Every experienced engineer who left took decades of compound knowledge with them. Most of it was never captured. The companies that recognised this early and built systems to capture, organise and query that knowledge will have an advantage that money alone cannot buy later. The window to capture it is not permanently open.
Four. OEMs will require data transparency across the supply chain.
Quality traceability, product carbon footprint, process records — the exchange of live production data between manufacturers and their customers will become a baseline requirement, not a differentiator. Suppliers who cannot provide it will not be suppliers for long. This is not a technology trend. It is a question of market access and licence to operate.
Five. Data access will be democratised (and that changes everything).
Every plant, every supplier, every operator will eventually have access to the same knowledge, the same models, the same tools. When that happens, having access to better data will not be enough. The advantage will belong to the companies that designed systems for speed before everyone else did. Organisations that build the right architecture early, to capture the knowledge, to deploy agentic AI, to establish trust between human and machine, will create a gap that competitors cannot close.
The direction is clear. The timing, as always, is the hard part.
AI won't evolve on its own
This is the part most AI commentary misses entirely.
The Industrial AI technology will improve faster than most sceptics expect. The number of mistakes will reduce, data will get cleaner, and integration will be faster. Anyone who is using today's imperfection as a reason to wait with Industrial AI is making the same mistake as the people who told Tercek his idea was ridiculous in 2000.
But the companies that build something genuinely useful won't just focus on understanding the technology. They will focus on understanding the human system it is entering, with all the fears, incentives and knowledge that lives in people's hands and retires with them.
How technology companies will approach this development will make a difference between a system that gets deployed and a system that gets used.
Design of the solution will determine whether the operator stops trusting an AI after the third false alarm, or whether the engineer quietly works around the system because nobody asked for his input when the solution was built.
Clarity of the ROI will determine whether the plant manager approves the investment or never understands what problem Industrial AI can actually solve.
Industrial AI technology alone won’t transform the organisations. People will, when they understand what is happening and why.
With this philosophy behind everything we build at Advanced Solutions, we not only focus on what the technology can do today, but also on where the structural logic of this
Industrial AI technology is taking us, and how humans and machines evolve together inside that process.
Tercek in 2000 was too early, but the direction was right.
We are also early now. The direction is right.
The question is not whether Industrial AI will reshape manufacturing. It will. The question is whether you will have built the architecture — the knowledge, the systems, the trust between human and machine — before the window closes.