Turn the stopwatch around
Every leader who commissions a time-and-motion study secretly hopes it will find the slow person.
The honest ones are prepared for it to find them.
Run one of these studies honestly and a pattern appears.
Trace twenty or thirty pieces of finished work back through their history. Don’t look first at where people worked hard. Look at where the work waited.
For a decision that needed one more meeting.
For an approval that was never going to be refused.
For a specialist who was already overloaded.
For a priority that changed on Tuesday and changed back on Friday, leaving half-finished work on everyone’s desk like coats nobody came back for.
In the factory, the bottleneck was often a machine.
In the modern organization, it is frequently a calendar.
And the calendar usually belongs to someone senior.
This isn’t an accusation.
It’s a design flaw.
We built organizations where authority concentrates and information flows upward, then act surprised when decisions queue at the top.
The AI paradox
AI is making one part of this problem much more visible.
It is making production cheaper.
Code can be generated faster.
Proposals can be drafted faster.
Analysis can be produced faster.
Documents can be summarized faster.
Agents can perform work that previously required a human to move between systems.
So we naturally ask:
How much more can we produce?
But there is another question we should ask first:
Who has to decide whether all of this production is worth accepting?
Because the cost of creating work is falling faster than the cost of judging work.
And that changes where the bottleneck lives.
The quiet casualties of the AI flood
There is a group rarely visible in AI adoption dashboards:
the reviewers.
The architect reviewing the code.
The partner signing off the proposal.
The finance lead approving the spend.
The product leader evaluating the recommendation.
The domain expert everyone sends their “quick question” to.
When drafts become almost free, these people don’t get more hours.
They get more queues.
Five times the drafts.
Five times the recommendations.
Five times the things that are “just one quick review.”
They respond the only ways humans can.
They skim.
They defer.
They rubber-stamp.
They work late.
Or they burn out.
Quality degrades quietly while the organization celebrates its increased output.
The AI didn’t necessarily fail.
The system did.
The human becomes the queue
This is where “human in the loop” can become a misleading comfort.
Suppose AI increases the amount of work entering a review process from 20 items to 100.
A human is still in the loop.
Technically.
But the human has become the bottleneck.
And once the queue grows large enough, the nature of judgment changes.
Instead of:
Understand → question → evaluate → decide
it becomes:
Scan → trust → approve → move on
This is particularly dangerous when the AI answer is plausible.
The obviously wrong answer gets challenged.
The polished, mostly-correct answer often doesn’t.
The organization hasn’t removed human judgment.
It has compressed it.
Turn the stopwatch around
Industrial management taught us to ask:
How fast is the worker?
The AI era needs a different question:
How long does the work take to move through the system?
Follow one piece of work.
A customer request.
A proposal.
A feature.
A support issue.
A decision.
Don’t begin by measuring how busy everyone was.
Measure:
Where did it wait?
The useful unit of analysis is the work item’s entire journey, from customer request to customer value realized, including touch time, waiting time, queue size, rework and decision ownership.
Because knowledge work often spends surprisingly little of its life actually being worked on.
The rest is waiting.
The worship of busy
We still confuse utilization with productivity.
A completely full calendar looks impressive.
But queueing theory tells us something uncomfortable: as utilization approaches 100%, waiting time rises sharply.
A team where everyone is fully booked is a team where every request waits.
So when I see a calendar with no white space, I no longer automatically see dedication.
I see a queue with a person’s name on it.
Slack isn’t laziness.
Slack is capacity.
Capacity to answer the question that unblocks three other people.
Capacity for a senior engineer to notice something nobody assigned.
Capacity for a manager to make a decision before the team spends another week waiting.
Capacity for a reviewer to actually review.
The goal isn’t to maximize how busy everyone is.
It’s to maximize how effectively the system flows.
Measure the work, never the worker
There is a dark twin to the stopwatch:
surveillance.
Keystroke counters.
Screen captures.
Activity scores.
Hours online.
They promise visibility.
Often, they produce the opposite.
People learn to perform being watched.
They look busy.
They hide the waits.
They stop telling you where the work is stuck.
And you lose the signal you needed most.
A better time study follows the work item, not the person.
Ask:
Where did this request wait?
And:
Who was it waiting for?
That question can be asked without shame.
Which is precisely why it can produce an honest answer.
The bottleneck may be leadership
This is the uncomfortable part.
Sometimes the constraint isn’t engineering.
It isn’t sales.
It isn’t operations.
It isn’t technology.
It’s decision latency.
One executive approves everything.
One person owns every exception.
Every important decision escalates upward.
Priorities change faster than teams can finish them.
Work gets started but rarely stopped.
Half-finished initiatives accumulate.
The organization has effectively built a queue around authority.
Then we ask why execution is slow.
Sometimes the answer isn’t:
“We need more people.”
It’s:
“We need fewer decisions to reach the top.”
Decision rights are a form of trust
One of the cheapest ways to widen a leadership bottleneck is to stop being one.
Write down which decisions you actually need to make.
Give everything else away with clear thresholds.
Below this amount:
You decide.
Below this risk level:
You decide.
Inside this scope:
You decide.
Only exceptions come to me.
This does two things.
It removes waiting from the system.
And it tells someone, in writing:
I trust your judgment.
That matters even more in an AI-enabled organization.
Because agents can eventually operate inside those same boundaries.
But the trust has to exist before the automation.
Otherwise we simply encode old organizational fear into new software.
The wrong return on AI
The most tempting story about AI is headcount reduction.
Sometimes that may be the right outcome.
But reducing people without understanding the constraint can remove exactly the slack, expertise and tacit knowledge the system depends on.
The cost line improves.
The system gets slower.
The customer waits longer.
A better question is:
Did we shorten the path from a customer’s need to a customer’s outcome?
Sometimes that means fewer people.
More often, it means the same people finally working on what matters instead of waiting for it.
The attached operating model makes the same distinction: the meaningful AI return is not adoption or generated volume, but improvement at the constraint and in end-to-end customer outcomes.
Where AI should actually go
AI shouldn’t automatically be pointed at the people producing the work.
First find the constraint.
Then ask what kind of constraint it is.
If the problem is handoff waiting, automate the handoff.
If the problem is review capacity, pre-review the work.
If the problem is decision latency, clarify decision rights.
If the problem is chasing approvals, let an agent handle the chasing.
If the problem is poor inputs, improve the quality gate before the bottleneck.
If the problem is too much work in progress, don’t generate more.
Stop starting.
The principle is simple:
Accelerate the constraint. Don’t accelerate everything.
Otherwise AI simply fills the queue faster.
The most important AI metric may be a queue
We love measuring:
AI adoption.
Agents deployed.
Tokens consumed.
Hours saved.
Outputs generated.
These numbers tell us that something happened.
They don’t tell us whether the organization got better.
Instead, measure:
Lead time.
Decision latency.
Review queue age.
Work in progress.
First-pass quality.
Accepted outcomes.
The goal isn’t more activity.
It’s faster movement from request to value.
What to do on Monday
None of this needs a transformation program.
It needs a leader willing to look.
1. Ask one question
“What are you waiting on from me right now?”
Write down every answer.
Clear what you can.
2. Count your work in progress
Including your own.
If you have ten strategic priorities, you probably don’t have ten priorities.
You have ten queues.
3. Find your reviewers
Name the three people whose “yes” the most work depends on.
Protect their capacity before adding another AI tool upstream of them.
4. Write down decision rights
What do you keep?
What do you delegate?
What are the thresholds?
5. Retire one vanity metric
Replace hours, utilization or AI adoption with something about the work:
How long does it take from request to customer value?
These are small interventions, but they attack the system rather than making individuals work harder.
The layer beneath the system
We talk about bottlenecks as if they are technical.
Underneath, they are often human.
Fear of letting go.
The comfort of being needed.
The habit of measuring what is easy instead of what is true.
The belief that a full calendar demonstrates leadership.
The belief that more output must mean more value.
AI is forcing us to confront these assumptions.
Because when machines can produce almost unlimited work, the scarce resource is increasingly not production.
It is:
attention.
judgment.
decision capacity.
trust.
And ultimately:
human responsibility.
The Human Layer Insight
The industrial era asked:
How do we make people move faster?
The agentic era asks a harder question:
How do we stop making people wait?
AI can make production almost limitless.
But it cannot make a decision for an organization that hasn’t decided who gets to decide.
It cannot make a reviewer absorb an infinite queue.
It cannot create trust by itself.
And it cannot tell a leader that their own calendar has become the bottleneck.
That part requires a human to look.
The answer usually starts in the mirror.
When did you last ask your team what they are waiting on from you?
Read the full issue on LinkedIn →
THE HUMAN LAYER publishes on LinkedIn. Subscribe here.