The Leadership Friction  |  Edition No. 26  |  August 25, 2026

The Human Challenge in an AI World  —  Part 2 of 8

Before You Automate It, Ask Whether It Should Exist

A faster process is not always a better one. Before any process gets automated, it needs to be diagnosed first.

AI makes it easier than ever to automate work.

Emails trigger themselves. Reports generate on a schedule. Meeting notes summarize without anyone touching them. Tasks assign automatically, and follow-ups happen whether or not a person remembers to send them.

That can save real time. I use these tools too.

But there is a question leaders need to ask before they automate anything, and most skip it entirely.

Should this process exist in the first place?

Because automating a bad process does not make it a good process. It just makes the bad process happen faster.

The Reframe

Efficiency Is Not the Same as Effectiveness

Organizations accumulate processes the way a house accumulates clutter. Slowly, and usually for a reason that made sense at the time.

A form gets added because something went wrong once. A meeting gets created because communication broke down. Another approval step gets added because someone made a poor decision three years ago and nobody wanted to risk it happening again. A report gets built because a leader wanted visibility into something.

Then nobody goes back and asks whether those steps are still necessary.

Years pass. Employees are working around processes nobody fully understands anymore, and the reason for the process has long since disappeared from institutional memory.

Then AI enters the picture.

Someone sees an opportunity to automate the process. That sounds like progress.

Automating a bad process does not make it a good process. It just makes the bad process happen faster.

But if the process is unnecessary, confusing, redundant, or built around a challenge that no longer exists, automation does not fix the dysfunction. It just makes the dysfunction less visible. The work moves faster. The organization is not necessarily working better.

Diagnose First

Start With the Purpose

Before you automate a process, ask a few basic questions.

Why does this process exist? What challenge was it originally designed to solve? Does that challenge still exist? Who actually needs the information or outcome it produces? What would happen if you stopped doing it?

Those questions can be uncomfortable, because the honest answer is sometimes, “We have always done it this way.”

That is not a strong enough reason to automate anything.

Look for the Friction First

When I look at a process inside an organization, I am not just looking at the steps. I am looking for where the friction is actually coming from, and it almost always shows up in one of three places.

Sometimes it is the system. Too many handoffs. Too many approvals. Too much duplication built up over years.

Sometimes it is communication. People do not know what information they need to provide, when they need to provide it, or who is supposed to receive it.

Sometimes it is people. Someone is holding onto a decision that should belong somewhere else. A manager does not trust the team enough to let go. An employee has never been trained properly. A leader keeps adding checkpoints because they are uncomfortable giving up control, not because the business needs the oversight.

Automate without understanding which of those three is actually creating the drag, and you will not remove the friction. You will just embed the challenge deeper into the organization, and build it into something harder to unwind later.

In Practice

Here Is What That Can Look Like

Imagine a company where every purchase over $500 requires three approvals.

As the company grows, that process becomes a genuine source of frustration. Purchases sit for days. Managers complain. Employees start looking for workarounds, which usually means the actual controls the process was supposed to provide are already gone.

Someone suggests using AI to route approvals faster. That could help.

But the better question is this: why do three people need to approve a $600 purchase in the first place?

Maybe they do. But maybe that process was built years ago, when the company was smaller and spending controls were genuinely weak. The organization does not need faster approvals. It needs clearer decision authority.

That is a very different solution, and no amount of automation gets you there.

The Sequence

AI Should Support the Way the Business Should Operate

There is a sequence that matters here, and most companies run it backward.

Decide how the work should happen. Clarify who owns what. Determine what information actually needs to move between people. Build the system around that. Then decide what AI can support.

Too many companies see what AI can automate first, and redesign the work around the technology second. That does not create simplicity. It creates a more sophisticated version of the same complexity, running faster than before.

Technology should support the operating model of the business. The operating model should not be built around whatever the newest technology happens to make possible this year.

The Leadership Responsibility

This is exactly why AI adoption cannot be handed off entirely to whoever knows the most about AI.

Leaders have to be in the room. They understand, or should understand, why the work happens the way it does. They should know where decisions actually belong. They are the ones positioned to see where unnecessary complexity has crept into the organization over time.

And they have to be willing to remove work, not just automate it.

Sometimes the best AI decision available to a leadership team is not to automate a process at all. It is to eliminate the process entirely.

One Question

Pick one process your organization is currently automating, or considering automating.

Before you talk about tools, prompts, or integrations, ask your team one question: if we were designing this process from scratch today, would we build it this way?

If the answer is no, do not automate it yet. Fix it first. Then let AI support a process that has actually earned the right to exist.