I think this is where some companies are getting a little too excited.
AI does one part of someone’s job really well and suddenly the conversation becomes:
“Why do we still need the person?”
That is a much bigger jump than it looks.
A task is not a job.
A support engineer does not just write summaries. They investigate, interpret, escalate, coordinate, judge severity, deal with uncertainty, speak to people and decide when the normal process is no longer enough.
AI may become very good at some of those things.
That still does not mean the whole role disappears.
1. A task is not a job
Most jobs are a collection of different types of work.
Some parts are repetitive.
Some parts are rule-based.
Some parts need judgement.
Some parts need context.
Some parts only become difficult when something unusual happens.
So when AI becomes good at one task, the right question is not:
“Can AI replace this person?”
It is:
Which parts of the work should still exist, and who or what should perform them?
That is a much more useful conversation.
2. Companies are automating before redesigning
This is another problem.
You start with a messy process.
Then someone says:
“Let’s add AI.”
Now you have the same messy process plus:
- an AI request;
- an AI response;
- a human review;
- some correction;
- another approval;
- and the same final update you had before.
Congratulations.
You have implemented AI.
The process may now be worse.
The better question is:
If we designed this workflow properly today, what would it look like?
That question comes before the AI.
3. Replacing capability too early can create a new problem
This is the bit I think organisations need to be careful with.
If AI removes a lot of easy work, that is great.
But the difficult cases do not disappear just because the easy ones became automated.
Eventually:
- the provider goes down;
- the model changes;
- the knowledge is wrong;
- a customer asks something unusual;
- the workflow hits an edge case;
- someone needs to make a judgement call.
If the organisation has already removed the people who understood the process deeply, the cost saving can turn into a dependency very quickly.
That does not mean organisations should never reduce roles.
It means the new operating model should prove itself first.
4. The real measure is not how much AI you use
I do not think “we use more AI now” tells us very much.
What actually matters is whether the work improved.
Did it get:
- faster;
- cheaper;
- easier;
- more reliable;
- less repetitive;
- more useful to the customer?
Did quality stay acceptable?
Did human review become smaller or did it quietly become a new job?
Did the process become simpler or more complicated?
Those are much better questions.
5. Sometimes AI is not even the answer
This one is easy to miss.
Some work should be automated with normal software.
Some work should remain human-led.
Some work should be supported by AI.
Some work should be performed by AI within clear boundaries.
And some work should probably disappear completely.
That is why I do not think the goal should be maximum AI.
The goal should be the best operating model for the work.
6. The five choices I keep coming back to
This became one of the central ideas in HAIOF.
For any piece of work, I think organisations should consider five possibilities:
Eliminate
The work should not exist.
Human-led
Human judgement or authority remains central.
AI-assisted
AI helps a person perform the work.
AI-executed
AI performs bounded work under clear authority.
Deterministic automation
Normal software or rules perform the work.
That is the conversation I think companies need to have before jumping from:
“AI can do this task”
to:
“We no longer need the person.”
The goal should be better work.
Not more AI.