The Great Automation Illusion
Every era of industrial technology has its favorite illusion. In the age of mechanization, it was the idea that machines would simply replace muscle. In the age of enterprise software, it was the belief that putting a process into a system meant the process was now better. In the age of dashboards, it was the assumption that seeing more numbers meant making better decisions. Now we are in the age of AI, and the illusion is even more tempting: if the work looks automated, then the work must have disappeared.
It usually has not disappeared. It has moved.
Sometimes it moves from the shop floor to the engineering team. Sometimes from the ERP system to a planner’s spreadsheet. Sometimes from the official process to the one person everyone calls because “they know how it actually works.” And sometimes, most dangerously, it moves into the human brain, where the extra work becomes invisible to everyone except the person stuck carrying it.
That is the great automation illusion. The company thinks it bought AI-driven operations, real-time decision-making, autonomous workflows, connected systems, and self-optimizing production. What is actually happening is often less impressive: operators manually correcting data, engineers babysitting integrations, Excel filling the gaps, tribal knowledge holding the process together, and humans acting as the real middleware.
The screen looks modern. The work behind it is very manual.
I do not say that as someone sitting on the sidelines yelling at the clouds. Having spent most of my career in sales, a portion of it in marketing, and more recently in strategy, I have seen a lot. I have seen the demo that looked amazing and the implementation that made everyone quietly miserable. I have seen systems that were technically live but operationally half-dead. I have seen dashboards that impressed executives and annoyed everyone who knew where the numbers came from. I have seen the same spreadsheet survive three transformation programs, two consultants, and a reorg.
Unfortunately, I don’t think leaders do not notice the difference between actual automation and human compensation because the business still appears to function. Orders ship. Meetings happen. Metrics get reviewed. Customers get answers. From far enough away, the operation looks fine.
Move closer, though, and the picture changes. You find people translating between systems that were supposedly integrated. You find engineers watching data flows like nervous parents at a playground. You find supervisors trusting the whiteboard more than the system. You find planners maintaining private files because the official tool is accurate in theory and useless in practice.
Automation Does Not Make Complexity Vanish
One mistake I see all the time is the belief that automation eliminates complexity. It rarely does. More often, automation moves complexity somewhere else. When companies do not deal with that complexity through better process design, cleaner data, clearer ownership, and more practical workflows, it lands on a person. Usually, a single very busy person.
A workflow gets digitized, but the inputs are still wrong. A machine sends signals, but nobody has agreed which signals matter. Two systems are connected, but nobody owns the data conflict between them. An AI tool makes a recommendation, but no one knows who is allowed to act on it. Then everyone wonders why the fancy system still requires so much manual intervention.
This is where manufacturing makes the problem painfully visible. Machines have quirks. Products have history. Quality issues have backstories. Maintenance decisions are full of tradeoffs. The official process tells part of the story, but anyone who has spent time around operations knows the unofficial process is often where the truth lives.
When companies automate without understanding that reality, they do not remove human judgment. They push it outside the system. People still make the process work, but now they do it through workarounds, side conversations, notes, spreadsheets, and memory.
That is a bad bargain. In a healthy operating model, human judgment is part of the design. People know when to step in, what they are responsible for, what data they can trust, and how exceptions get handled. In a weak operating model, judgment becomes glue. It holds everything together, but nobody documents it, measures it, funds it, or respects it until the person who knows everything goes on vacation. And then all of a sudden everyone cares.
The Human Brain Is Still the Middleware
The line at the bottom of the graphic was actually the last thing I added: “The most overworked integration platform is still the human brain.” I added it because without that line, the graphic felt like it was only poking fun at bad automation. And yes, I am absolutely poking fun at it. But the bigger point is more serious.
I have seen this from too many angles now. In sales, it shows up when the system cannot capture the real nuance of the customer conversation, so the rep carries it in their head. In marketing, it shows up when the dashboard says one thing, the campaign data says another, and someone has to explain why both are “technically right.” And now in strategy, I see it in a different form: leaders want AI, automation, and faster decisions, but the business still runs on people quietly translating between systems that were supposed to make that translation unnecessary. That is why the quote mattered to me.
If your automation still depends on people constantly correcting, remembering, interpreting, and stitching things together behind the scenes, then you did not remove the work. You just made it harder to see. And AI will not clean that up by itself. AI needs context. It needs meaning. It needs a business that understands its own decisions. Otherwise, we are not making the operation smarter. We are just feeding confusion into a faster machine. Before we ask AI to think for the business, we should probably stop making humans compensate for systems that still cannot understand each other.
Connected Is Not the Same as Useful
Companies love saying their systems are connected. I get why. It sounds like progress. It looks good in a strategy deck. It gives everyone the warm feeling that data is flowing majestically across the enterprise.
But connected does not always mean useful. Connected means data can move. Useful means the right data can be trusted by the right person at the right moment to make the right decision. Those are not the same thing. A historian, MES, ERP, quality system, maintenance system, and planning tool can all be connected and still produce confusion if no one has defined ownership, meaning, priority, and action.
This is why “real time” can also be a trap. Real-time data is only helpful if the organization can do something with it in real time. Otherwise, it is just faster noise. A production alert is not valuable because it appears instantly. It is valuable because someone understands what it means, what choices exist, what constraints matter, and what action should follow. The same applies to AI recommendations. An answer is not useful simply because it arrives quickly. It is useful when the recommendation fits the context, respects the constraints, and enters a process where someone knows what to do with it.
That is where many companies have work to do. A lot of work…
What Leaders Should Do Differently
wanted to include this section because it is too easy to make fun of the problem and stop there. The harder, more useful move is to say what leaders can actually do about it. My hope is that this gives people a way to look at their own automation efforts and ask, “Are we removing friction, or are we just hiding it better?”
First, map the work people are doing outside the system.
In almost every company I have worked with or around, the official process and the real process are not the same thing. The real process is in the spreadsheet someone refuses to delete, the side conversation before the meeting, the manual check before the report goes out, and the person everyone calls when the system answer feels wrong. Leaders should actively look for those places, not shame them. Ask teams: “Where do you still have to double-check the system? Where do you keep your own version of the truth? Where do you fix something before anyone else sees it?” That is where the real automation opportunity usually starts.
Second, find out why the workaround exists before trying to eliminate it.
I have seen leaders get frustrated when teams still use Excel, email, whiteboards, or informal trackers after a new system goes live. I get it. It looks like resistance. But in my experience, the workaround is usually telling you something important. Maybe the system is too slow. Maybe the data is not trusted. Maybe the workflow does not match how decisions actually happen. Maybe the tool was designed for reporting up, not helping people do the work. Before banning the workaround, study it. The workaround may be the most honest documentation you have.
Third, be painfully clear about which system owns which truth
One of the fastest ways to create chaos is to have five systems all pretending to be authoritative. The ERP says one thing, the MES says another, the quality system adds nuance, the maintenance system has history, and the planner’s spreadsheet somehow knows what everyone is actually going to do. I have seen this turn simple decisions into meetings about whose number is “right.” Leaders need to define this explicitly: which system owns the customer order, production status, quality disposition, inventory availability, asset condition, and schedule commitment? When systems disagree, the answer should be governed, not negotiated by whoever argues best.
Fourth, separate automation from decision-making.
This is where companies get ahead of themselves. Automating a task is one thing. Automating a decision, or even recommending one, is much more serious. Before dropping AI into a workflow, leaders should ask: What exact decision are we improving? Who owns it today? What information do they use? What tradeoffs matter? What should the system do when confidence is low? What should be escalated to a human? I have become a big believer that “the AI should stay quiet here” is sometimes one of the smartest design decisions you can make.
Fifth, stop rewarding heroics as if they prove the system is healthy.
This one is personal because I have seen too many companies celebrate the person who saves the quarter, fixes the report, reconciles the mess, or keeps the customer calm because the process failed again. That person deserves recognition, but the system also deserves scrutiny. If the same person keeps saving the day, that is not just talent. That is a design flaw with a very reliable employee attached to it. Thank them, then fix the reason they had to be heroic in the first place.
Sixth, design the human role intentionally.
The goal is not to remove people from every process. That is lazy automation thinking. The goal is to remove people from the work they should not have to do: duplicate entry, manual reconciliation, chasing approvals, translating between systems, and babysitting bad handoffs. Then be explicit about where humans matter most: judgment, exception handling, customer nuance, ethical tradeoffs, prioritization, and improvement. In my opinion, the best automation does not make people disappear. It makes their contribution more obvious, more valuable, and less buried under administrative nonsense.
