The logic of automation is clean, which is part of why it is dangerous. You look at a role, you break it into the tasks the person performs, you find that a tool can now do several of them, and you conclude that you can remove that share of the cost while keeping the work. The role becomes a list. The list gets shorter. The savings look obvious.

The problem is a quiet assumption buried inside that logic: that a role's value and a role's tasks are the same thing. Sometimes they are. Often they are not, and the difference is invisible in exactly the way that matters, because automation can only see the tasks. It cannot see the value that was never written into the task list in the first place. Which means it is entirely possible to automate everything a role visibly does and hollow out most of what it was actually worth, while the org chart still looks intact and the savings still show up on the spreadsheet.

The value that was never in the job description

Think about your best regional manager, or your most trusted person in accounting, or the maintenance lead everyone calls. If you wrote down what they do all day, you would capture the tasks: the reports, the approvals, the calls, the entries. What you would miss is most of why they are valuable.

You would miss the judgment, the sense of when a number looks wrong before anyone can say why, when a vendor is about to become a problem, when a tenant situation is going to escalate. You would miss the exceptions they catch, the small things that would have become large things if a less experienced person had not quietly handled them. You would miss the relationships, the reason a contractor picks up their call, the reason an owner trusts the statement without auditing it line by line. And you would miss the tacit knowledge, the accumulated feel for how this specific operation actually behaves, which lives in a person and appears in no document.

None of that is in the task list. All of it is in the value. When you automate the tasks and remove the person, the tasks get done and the value walks out the door, and because the value was never measured, its absence does not show up until something that should have been caught is not.

Why automation is worst at exactly this part

This is not a coincidence, and there is a well-established principle in AI research that explains it. It is called Moravec's paradox, articulated in the 1980s by robotics researcher Hans Moravec along with Rodney Brooks and Marvin Minsky, and it observes something counterintuitive: the tasks that are hard for humans, structured calculation, formal reasoning, processing large amounts of data, are often easy to automate, while the tasks that are easy for humans, judgment, physical intuition, reading a situation, sensing that something is off, are extraordinarily hard to automate. We have it backwards from how it feels. The impressive-looking cognitive work is the automatable part. The effortless-seeming human part is the stubborn one.

Apply that to a role and the danger becomes precise. The parts of a job that are easiest to automate are the codified, visible tasks, the exact things on the task list. The parts that are hardest to automate are the judgment and the situational sense, the exact things that made the person valuable and that never made it onto the list. So automation does not just happen to miss the valuable part. It systematically removes the easy, visible part and leaves you holding a role stripped of its tasks but still responsible for the judgment, now with no one doing it. You have not made the role more efficient. You have separated its work from its worth and kept only the work.

The trap that shows up later

There is a second-order version of this that is easy to miss because it operates on a delay. The routine tasks a role performs are not only output. They are also how judgment gets built. The analyst who formats the data learns to see when it is wrong. The junior person who handles the routine cases develops the instinct for the hard ones. The tacit expertise you value in your senior people was manufactured, over years, by doing exactly the unglamorous tasks you are now tempted to automate away.

This effect is well documented, most clearly in aviation. NASA research has found that pilots who lean heavily on automation see their cognitive flying skills erode, the ability to track the aircraft's position and situation without the automated systems, even as other skills hold up. The same pattern now appears in knowledge work: a peer-reviewed review of AI overdependence documents how sustained reliance on automated tools drives skill atrophy, the degradation of expertise that simply stops being practiced. So automating the visible tasks can quietly starve the pipeline that produced the invisible value in the first place. You save on the tasks now, and you find in a few years that the next generation of judgment never formed, because the reps that would have built it were handed to a tool. The cost is real, it is just deferred past the quarter where the savings were booked.

The honest part: plenty of roles are their tasks

It would be wrong to turn this into an argument against automation, because the whole point is that the distinction is what matters, not a blanket verdict. Plenty of roles genuinely are their tasks. Work that is purely transactional, where the value is the accurate completion of a defined process and there is no meaningful layer of judgment, relationship, or exception-handling underneath it, is exactly where automation belongs. Automating that work is not hollowing anything out. It is freeing people from drudgery that was not building much of anything, and refusing to do it out of vague sentiment would just be waste dressed as caution.

So the goal is not to protect every role from automation. It is to stop treating every role as if it were only its task list, and to do the harder work of telling the two kinds apart before you cut.

The question a COO actually has to answer

For any role you are considering automating, the discriminating question is not "can a tool do these tasks." A tool can do more tasks every year, so that question always trends toward yes and tells you almost nothing. The question is: if this role's visible tasks were fully automated tomorrow, what would stop happening that no one is currently measuring? What judgment would go unexercised, what exceptions would go uncaught, what relationships would go untended, what would we only discover we had lost six months later when something that used to get handled quietly becomes a problem loudly?

If the honest answer is "nothing much, the tasks are the job," automate with confidence. If the honest answer is a list of things you had never written down because they were simply expected to happen, you have just found the value you were about to delete by accident. The discipline is to ask the question deliberately, for each role, rather than letting the task list answer it for you by default, because the task list will always say the role is safe to automate. The task list is the one thing in this decision that cannot see what actually matters.

FAQs

Q1. Isn't this just an argument to slow down on automation?
No. It is an argument to automate the right things, which requires separating roles that are genuinely their tasks from roles whose value lives in judgment and relationships the task list never captured. Slowing down uniformly would waste the real gains available in transactional work. The point is precision, not caution.

Q2. How do I tell whether a role's value is in its tasks or beyond them?
Ask what would quietly stop happening if the visible tasks were automated and the person left. If the answer is essentially nothing, the value is in the tasks. If the answer is a set of judgment calls, caught exceptions, and relationships you had never formally documented, the value lives beyond the tasks and is at risk of being removed without anyone noticing.

Q3. What is Moravec's paradox and why does it matter here?
It is the observation from AI research that the work humans find hard, formal calculation and data processing, is often easy to automate, while the work humans find easy, judgment and situational sense, is very hard to automate. It matters because it means automation naturally removes the visible, listed tasks and leaves the unlisted judgment undone, which is the exact inversion of what you want.

Q4. What is the deskilling risk you mention?
Routine tasks are not only output, they are also how people build judgment over time. When those tasks are automated away, the experience that used to produce your senior people's expertise stops accumulating. The immediate efficiency can therefore starve the long-term pipeline of judgment, a cost that shows up years later rather than in the quarter you automated.

Q5. Doesn't keeping humans in roles that could be automated just protect inefficiency?
Only if the role's value really was just its tasks, in which case yes, automate it. The argument is not to keep people in transactional roles out of sentiment. It is to recognize that some roles carry substantial unmeasured value, and removing them looks like efficiency on the spreadsheet while actually destroying something the spreadsheet could not see.

Q6. How can I measure value that was never written down?
You often cannot measure it directly, which is the core difficulty, but you can surface it by asking the people around a role what that person catches, handles, or prevents that no one formally tracks. The informal answers, who relies on them and for what, tend to reveal the judgment and relationships that the job description omitted.

Q7. Where is automation clearly the right call?
In genuinely transactional work, where the value is the accurate execution of a defined process and there is no meaningful layer of judgment, exception-handling, or relationship underneath. That work is where automation delivers clean gains, and hesitating there just leaves value on the table. The skill is distinguishing it from work that only looks transactional on paper.

Q8. What is the first practical step before automating a role?
Write down not just the role's tasks but what would go uncaught or unmanaged if those tasks were automated and the person left. That second list, the one that usually was never made, is what tells you whether you are removing cost or quietly removing value. Making it explicit before you decide is the whole discipline.