AI can generate a workflow in seconds. The difficult part is knowing whether that workflow actually makes sense in the organisation.
This is one of the things I have become much more interested in as AI has moved into everyday business work.
Give AI a prompt such as "create a document approval workflow for a regulated organisation" and you will get something that looks convincing almost immediately.
There will be stages. There will be owners. There will be approvals, notifications and escalation points.
It can look finished.
But a workflow is not just a diagram.
It is an agreement about who does what, who is allowed to make a decision, what information is used, what happens when something goes wrong, and what evidence is left behind.
That is where things become much more complicated.
A workflow can look right and still be wrong
A simple workflow might look like:
Create → Review → Approve → Publish → Archive
Logically, there is nothing wrong with it.
But real organisations are rarely that simple.
Who creates the document?
Who reviews it?
Can the reviewer also approve it?
What happens if the approver is away?
Where is the controlled version stored?
What happens to the previous version?
Does anyone need training after the change?
What happens if someone is still using a downloaded copy?
These are not questions AI can answer simply because it has generated a nice-looking workflow.
They require knowledge of how the organisation actually operates.
AI sees what we give it
This is probably the most important limitation to understand.
AI can work with the information we provide. It can identify patterns, suggest structures and produce a very good first version.
But it does not automatically know what is happening behind the process.
It does not know that the person listed as the process owner left the company six months ago.
It does not know that employees have started saving documents somewhere else because the official location is difficult to use.
It does not know that the procedure exists but nobody actually follows it.
And it does not know what an auditor will ask for six months from now.
That information comes from people and from operational experience.
The dangerous part is that the output looks professional
This is where I think AI introduces a different kind of risk.
A bad workflow is easy to reject.
A plausible workflow is much harder.
You can look at it and think: "This makes sense."
But something can be logically correct and operationally wrong.
For example, an automated document approval process might work perfectly from a technical perspective.
The request is submitted.
The approval goes to the right person.
The document is published.
The workflow closes.
Technically, everything worked.
But what if the wrong version was submitted?
What if the employee who needs the new procedure was never trained?
What if an uncontrolled copy is still being used?
What if the person approving the document does not actually have the right authority?
The automation worked.
The control system did not.
This is where governance matters
When I think about workflow design, I am interested in much more than the sequence of steps.
I want to know:
Who owns the process?
Who owns the decision?
What information is required?
Who has authority?
What happens when the normal process does not apply?
What evidence needs to exist afterwards?
Who checks that the process is still working?
Those questions are what turn a workflow into an operational system.
Without them, we are mostly drawing boxes.
I don't think this means we should use AI less
Actually, I think the opposite.
AI can be extremely useful in this work.
It can help create the first version of a workflow much faster.
It can compare procedures.
It can identify missing steps.
It can generate questions for process discovery.
It can help analyse large amounts of documentation.
It can suggest alternative process structures.
That is valuable.
But I would treat the output as a starting point, not as the final answer.
The model I keep coming back to is simple:
AI generates.
People challenge it.
The organisation validates it.
Governance controls it.
The question I would ask instead
Instead of asking:
"Can AI create this workflow?"
I think the better question is:
"What would need to be true for this workflow to work safely in the real organisation?"
That question immediately changes the conversation.
Now we are talking about ownership, permissions, data, exceptions, training, adoption and accountability.
And that is where the real work starts.
The more capable AI becomes at producing operational content, the more important human review becomes.
Not because AI is useless.
Because good operational design has always depended on context.
The opportunity is not to spend hours drawing the first workflow anymore.
The opportunity is to use that saved time to ask better questions about whether the workflow actually reflects reality.
That is where I think the human value is moving.
Less time creating the first version.
More time challenging whether the first version is actually right.
AI can help us build the workflow.
It still needs someone to ask whether the organisation can live with it.