AI Can Write Our SOP. It Can't Tell Us What Our Process Should Be.

A few days ago, I opened Upwork.

I wasn't actually looking for a job or planning to hire a freelancer. I was just curious.

With AI becoming part of everyday business operations, I wanted to see what kinds of projects businesses were still hiring people to handle.

What I found was interesting.

I kept seeing requests for experienced business operations specialists, SOP writers, workflow designers, and process-improvement professionals.

Some businesses wanted someone to document their procedures. Others needed help organizing workflows, improving operations, or figuring out how their work should be structured.

And I found myself wondering:

With all the AI tools businesses now have access to, why are they still looking for people to do this work?

The more I thought about it, the answer seemed fairly simple.

Having the tools isn't the same as understanding the work.

AI can produce an SOP. But what should the SOP say?

Give an AI tool enough information about a recurring activity and it can help turn that information into a remarkably polished procedure.

It can organize steps.

It can improve the language.

It can create checklists, tables, role descriptions, templates, and training materials.

That is genuinely valuable.

I use AI myself, and I wouldn't want to go back to working without it.

But there is a question that comes before all of that:

Is the process we're asking AI to document actually the process we should be using?

Imagine a workflow that contains unnecessary approvals, duplicate data entry, unclear responsibilities, workarounds, inconsistent handoffs, and steps that have survived for years simply because nobody has questioned them.

AI can document that workflow beautifully.

But we may end up with a beautifully written SOP for a process that shouldn't exist in its current form.

Before documentation comes operational understanding

This is where I think we sometimes overestimate what having an AI tool actually gives us.

Before we can create a useful procedure, someone has to understand the operation well enough to ask questions such as:

What actually happens today?

What is supposed to happen?

Why does this step exist?

Who owns the work at each stage?

Where does information come from?

Where does responsibility change hands?

Which exceptions actually matter?

Which activities add value?

Which steps should be simplified?

Which steps should disappear completely?

And eventually:

Which parts of this work should we automate?

These aren't primarily writing questions.

They are operational questions.

And answering them often requires something AI doesn't automatically receive when we open a prompt box: context.

The difficult part is often not creating the document

An experienced employee may tell you, “This is how we process the request.”

But when you observe the work, you may discover that the written procedure hasn't been followed in two years.

A manager may tell you that one department owns a particular step.

The employees doing the work may tell you that responsibility actually moves back and forth between three people.

A process map may show five steps.

In reality, employees may be using spreadsheets, emails, side conversations, personal reminders, and undocumented workarounds just to keep those five steps moving.

That operational reality matters.

Someone has to recognize it.

Someone has to ask why it is happening.

And someone has to decide what the better way of working should look like.

Only then does documenting the process become truly useful.

AI doesn't make operational knowledge less important

In some ways, I think it makes operational knowledge more important.

Businesses are increasingly able to automate tasks that previously required significant manual effort.

AI can analyze information, draft communications, summarize documents, generate reports, monitor activities, and support decisions.

AI agents will be able to perform increasingly complex sequences of work.

But the more work we give technology, the more clearly we need to define the work we're giving it.

If responsibility is unclear between two employees, adding automation doesn't automatically clarify the responsibility.

If a workflow contains unnecessary steps, automating them doesn't make those steps necessary.

If nobody agrees on what should happen when an exception occurs, an AI system still needs rules, context, or a way to escalate the decision.

And if the organization itself doesn't understand how the work should flow, buying another AI tool may simply give it another powerful tool without a clear operating model.

Some operations work will change

I don't think the right conclusion is that AI cannot replace anything operations professionals currently do.

It can.

And it will.

Drafting procedures, formatting documentation, analyzing operational data, producing routine reports, monitoring activities, and many administrative parts of operations work can increasingly be accelerated or performed by AI.

That changes what valuable operational expertise looks like.

The value begins moving upstream.

Less time may be needed to physically create the SOP.

More value may come from knowing whether that SOP represents a good process in the first place.

Less time may be spent manually assembling a report.

More value may come from determining what should be measured and why.

Less time may be spent moving information between systems.

More value may come from designing how information should move through the organization.

That isn't resisting AI.

It's using it where it actually creates value.

The tool and the operational mind

Looking through those Upwork projects reminded me of something that is easy to overlook when we talk about AI replacing work.

Businesses don't simply need documents.

They need functioning operations.

They need someone to recognize why work keeps getting stuck.

Someone has to notice when a procedure doesn't match reality.

Someone has to determine whether the problem is the task, the handoff, the ownership, the information, the system—or the process itself.

Someone has to decide what should change before asking technology to execute the change.

AI can be extraordinarily useful in doing that work.

But giving someone an AI tool doesn't automatically give them operational judgment.

Just as owning sophisticated accounting software doesn't make someone an accountant, having powerful workflow and AI tools doesn't automatically tell an organization how its operations should be designed.

The opportunity, then, isn't to compete with the technology.

It's to become better at the part of the work where judgment matters—and use the technology to make everything around that judgment faster and stronger.

The future of operations isn't people versus AI.

It's people who understand the work using AI to improve how the work gets done.