What Happens When We Automate the Pause?

We spend an extraordinary amount of time trying to make work faster.

We streamline processes. Eliminate unnecessary steps. Reduce handoffs. Automate repetitive tasks. And increasingly, we introduce AI agents capable of completing in seconds what once took people hours to do.

Much of that is progress, undeniably so. But I've been thinking about something we may be eliminating along the way - something we rarely recognized as valuable in the first place.

What happens when we automate the pause?

Not the scheduled approval. Not the formal control. Not the obligatory sign-off someone added to a workflow years ago.

I'm talking about the figurative speed bumps that existed simply because a person was doing the work.

→  The moment someone noticed that a number looked different than expected.

→  The question that emerged while reconciling two conflicting reports.

→  The experienced employee who recognized something familiar about an unusual transaction.

Or the junior analyst who asked a question that made everyone reconsider an assumption they'd long ago stopped questioning.

Those moments weren't necessarily documented. They may never have appeared on a process map. And they certainly weren't all valuable.

But some were.

And when we automate the work, do we fully understand what else, perhaps inadvertently so, has been removed?

Have you been wrestling with this, too? I’d love to compare notes.

I've been turning this question over from a few different angles - what happens to judgment, how people develop expertise, and whether automation might create opportunities we haven't yet imagined.

Here are some of the places my thinking has taken me.

Friction Isn't Always Just Friction

Let's be clear: I'm not advocating for preserving inefficient processes.

I've spent much of my career helping organizations improve how information moves and decisions get made. I know how much time unnecessary manual effort consumes.

But there's a difference between eliminating a task and eliminating something valuable that happened while the task was being performed.

Consider a financial analyst preparing a forecast.

They gather information, reconcile inconsistencies, and review assumptions. Along the way, they may notice a changed definition, a broken historical relationship, or technically accurate information being used for the wrong decision.

Now imagine an agent completes that same workflow in seconds.

The forecast may be more accurate and consistent than what the analyst would have produced.

Wonderful.

But where did the opportunity to question the assumptions go?

Maybe it's embedded in the agent's instructions. Maybe automated tests identify anomalies. Maybe another agent independently challenges the result.

Or maybe nobody thought to ask.

The issue isn't how quickly the work gets done. It's whether the judgment that matters still has somewhere to happen.

The Human in the Loop Isn't Automatically the Answer

It's tempting to respond by insisting that a human remains involved. But that begs us to contemplate another question: involved in what, exactly?

If someone reviews an agent's output without understanding the underlying information, assumptions, or consequences, have we preserved meaningful judgment? Or have we simply introduced another approval?

And what happens when the agent is consistently more accurate than the person reviewing it?

Perhaps the human needs more training. Perhaps management needs to reassess decision rights - or better yet consider where human intervention adds (the most) value and shift accordingly.

Both of the following scenarios can be true simultaneously:

→  The presence of a person doesn't guarantee the presence of good judgment

→  The use of automation doesn’t necessarily eliminate good judgment

And that's precisely why the design matters.

We have opportunities to design a state of ‘challenge’ directly into automated workflows.

An agent might flag uncertainty or information outside established tolerances. Automated tests could detect unexpected changes. Another agent could independently evaluate selected assignments, with material differences triggering investigation.

For consequential decisions, an agent might be required to stop when essential context is missing. For lower-risk activities, it might proceed while clearly identifying its assumptions and limitations.

These mechanisms have limits. Two agents can agree and still be wrong. Thresholds, just like human reviewers, can miss problems.

We needn’t recreate the old pause frame for frame. We need to understand its purpose and decide how best to fulfill it. And the answer won't be the same for every organization, workflow, or decision.

What Happens to the People Who Learned by Doing?

What happens to the development of expertise when the work through which people traditionally acquired that expertise becomes automated?

I spent years at Bloomberg working with financial information, including corporate bonds and credit analysis. I can attest to the axiom of, “Professional competence isn’t simply a function of tenure.” Why? Because the conditions we experience and build professional competence through matter.

An analyst working through a relatively benign credit environment may develop strong technical skills without encountering the complexity of covenant issues that emerge during a tightening credit cycle.

They might understand the inverse relationship between bond prices and yields. But have they experienced a flight to safety? The panic at the desk when supposedly “secured” debt turns out to be secured by equity in a subsidiary with no assets to sell?

Can they navigate a conversation with a stressed trader? Explain why the data behaves as it’s been coded to behave, even when that conflicts with how someone expects it to behave?

Knowing the theory is one thing. Exercising judgment when the conditions change is another.

Historical examples and simulations can help, but there's tremendous value in experiencing how decisions, risks, and consequences unfold in real operating conditions.

If we automate the foundational work, how will the next generation develop that understanding? I don't think the answer is to make junior employees perform repetitive tasks indefinitely.

Perhaps they need deliberate apprenticeships combining practical experience, classroom learning, and increasing responsibility for training, testing, and assessing agents. The duration might depend on the complexity of the work, the relevant business cycles, and the conditions they've actually experienced.

Changing market conditions, regulations, or agent capabilities might also trigger further education and reassessment.

We already have useful precedents from which we can draw upon. My own Certified Data Management Professional (CDMP) qualification offers progressive levels of recognition. I have it on a few very good authorities that the accounting and legal worlds have developed their own continuing education requirements.

Why wouldn't we borrow from what's already working?

Perhaps we can adapt structured apprenticeships, progressive qualifications, and continuing education for people responsible for automated work.

Professional associations, universities, and independent providers could share that responsibility with employers.

Why invent a new wheel when we can borrow what works and improve upon it? Wouldn’t that get to the very heart where continuous improvement meets continuous learning?

What If the Student Becomes the Teacher?

Imagine a junior practitioner and an experienced subject matter expert independently configuring agents to perform the same type of analysis. Both agents are then given identical scenarios, and their outputs are evaluated not only for accuracy, but also for where and why their approaches differ.

The junior's agent might miss a subtle exception that the experienced practitioner anticipated. That's an opportunity for learning. But what if the junior's agent approaches the problem differently and identifies an assumption the expert has stopped questioning? Now the student has something to teach.

And what if neither approach is quite right, but comparing them reveals a third possibility that neither person had considered? That's something more than apprenticeship. It's the creation of new knowledge.

Of course, we'd need independent evaluation - not an assumption that the expert is always right or that novelty is inherently better. The junior develops judgment by articulating and testing their understanding. The expert revisits assumptions that experience may have made habitual. And the organization sees differences in reasoning that might otherwise remain hidden. Now that is cause for celebration, right?

Perhaps automation could also make some of the judgment that once depended on individual experience more visible, teachable, and open to challenge.

The automation could help us develop expertise, rather than simply reduce our perceived dependence on it. Apprenticeship usually means helping the next generation learn from established experts. What if it also helped experts keep learning from those coming behind them? Could take some of the pressure off of always feeling replaceable - by age, or machine…

Judgment Needs Somewhere to Live

Deploying an agent doesn’t make any of this happen automatically.

Someone still needs to determine what the agent is authorized to do, what constitutes an acceptable result, when intervention is required, and who is accountable when something goes wrong.

Believe me, I don't think the answer is necessarily another centralized committee or a new executive title.

We already have models for distributing responsibility. In data governance, for example, federated structures can allow routine issues to be handled close to the work, more consequential matters to escalate through domain leadership, and enterprise-level risks to reach the appropriate decision-makers.

The same principle could apply to autonomous workflows. The people who understand the work, the people accountable for its outcomes, and the people affected by those outcomes should help determine where judgment is needed. Depending on the circumstances, that might involve business leadership, technology, risk, HR, legal, compliance, or client-facing teams.

Someone needs both the authority and the expectation to act, and the organization needs a way to recognize when the existing approach is no longer sufficient. However, clear authority is only part of the equation. What happens when we build opportunities to question an agent into the workflow, while rewarding everyone for uninterrupted execution?

Imagine an employee who stops an automated process because something doesn’t feel right. They investigate, challenge an assumption, and discover a problem before it reaches a client. From one perspective, they prevented a costly mistake. From another, they interrupted throughput and missed a productivity target.

Which behavior will the organization celebrate?

Or consider the reverse: a team reports fewer escalations and faster turnaround. Is that evidence of better automation - or have people learned that raising a concern creates more trouble than it’s worth?

We can design the most thoughtful checks and escalation paths imaginable. But if the surrounding culture discourages their use, judgment still has nowhere meaningful to live.

This is where leadership, incentives, and measures of success become part of the design. What do we recognize? What do we reward? Do people have permission - and enough trust - to say, “Yes, the agent can do this. But should we?”

I don’t think we need to know the perfect answer before we begin. We do need to be willing to notice when the design isn’t working, learn from it, and change course.

We also need evidence that people are exercising judgment rather than passively accepting an agent’s output. Are they challenging outputs, investigating discrepancies, improving controls, escalating concerns, or returning to training when conditions change? Those activities show engagement, not necessarily better judgment. We’d also need evidence that interventions work and business outcomes improve.

Because a human in the loop who has stopped thinking isn't much of a safeguard.

Perhaps We Can Make the Pause More Valuable

Not every manual process contains hidden treasure. Some are simply inefficient.

And not every automated workflow needs a human reviewing each decision.

But we should understand what changes when people stop executing the work - and begin overseeing, questioning, teaching, and improving it.

→  What happens to judgment?

→  What happens to expertise?

→  What happens to the next generation who would have learned by doing?

And what might become possible if we deliberately design for challenge, learning, accountability - and a culture that actually welcomes them?

Organizations will start from different places. What works will depend on the work itself, organizational maturity, and the consequences of getting things wrong. Some experiments will fail. Some safeguards will prove unnecessary. Some questions will lead us somewhere we never anticipated.

That’s not a reason to wait until we have every answer. It’s a reason to approach this with intention: test, observe, challenge, learn, and refine - together.

Where does this leave us? Not with a mandate to slow automation, but with a bigger ambition: to grow our capacity to question, learn, and decide alongside our capacity to execute.

Before automating the next workflow, perhaps we start by asking the people who perform it:

→  What do you notice, question, or correct along the way that isn’t captured in the process documentation?

And then ask the more ambitious question:

→  If people no longer spent their time performing this work, what could we enable them to notice, question, learn, or improve instead?

Organizations that get good at this may do more than protect the judgment they once relied upon. They may discover better decisions, stronger practitioners, more productive challenges - and unexpected value that routine execution never left room to explore.

The pause was never valuable simply because it slowed us down. It was valuable when it made room for judgment.

We don't need to position human judgment and AI capability as competitors. We have an opportunity to design how they work together into our everyday operating practices - so that judgment isn't simply preserved, but developed, exercised, and celebrated.

Perhaps the organizations that master this won’t simply automate work more effectively. They’ll discover new ways for people and technology to think, learn, and create value together.