“Son, don’t go into banking. Do something sensible. Become a plumber.”

Imagine Mum saying that.

For decades, the office was where you went for a respectable career. Clean hands. A desk. Perhaps, eventually, people reporting to you.

Now I’m wondering whose chair starts wobbling first.

My bet? Look closely at middle management.

But first, let’s visit Grandma Gretchen.

She’s sitting in her apartment, wrapped in a scarf, wondering why the heating isn’t working.

Across the street, Mr and Mrs Schlabotnik are watching their automated awning extend and retract at two in the morning.

Their smart-home panels glow reassuringly green.

Every system works perfectly on its own. Together, they’re idiots.

Somebody has to turn up. Understand what is happening. Find the faulty sensor, the stuck valve or the conflicting instructions. Get the whole thing working.

Grandma Gretchen needs a warm apartment. The Schlabotniks need some sleep.

That takes more than a dashboard.

The physical world remains stubbornly physical. Even when AI helps diagnose the problem, someone still has to reach the valve.

Meanwhile, back at the office, we have people collecting information, moving it between systems, preparing reports, chasing approvals and asking other people for updates.

Then we have people managing the people doing all that.

That is where I expect the chairs to wobble.

Here is my deliberately steep hypothesis:

In an office built around repetitive administrative work, agents could perform 80% of the standard tasks better and faster.

A hypothesis. Put it to the test. Count the errors, the exceptions, the supervision and the rework.

Give me two people of my choice and eight weeks. Let’s find out how much of that work we can redesign — and how much still needs human judgment.

Because once agents can do the collecting, comparing, routing and reporting, what exactly are we managing?

If your contribution is primarily distributing standard tasks and consolidating the results, you should be asking that question now.

You will have to get involved in solving the actual problem. Understand the work. Make decisions. Take responsibility for outcomes.

And yes, there will be resistance.

There always is when the thing being redesigned is somebody’s position.

I recognise this from the enterprise architecture world. At some point, drawing the architecture has to become responsibility for a product that works.

The same challenge now faces management.

For me, this is product management: understand what people need, shape the solution, build the agents with the team and own whether the result is useful.

Automating a pointless process gives you a faster pointless process.

Someone has to decide what should exist in the first place.

And the Product Manager gets no permanent exemption.

You must plan for the possibility that your own role becomes unnecessary.

If the product only works while you personally coordinate every move, you may have built yourself into the problem.

That applies to me too.

I expect this change to arrive sooner than many organisations are prepared for. Waiting for everyone to feel comfortable is a poor plan.

But treating people as disposable is an equally poor one.

People need a real opportunity to learn, contribute and take on new responsibility. Leadership has to make that possible — and people have to engage with it.

Respect does not require us to preserve every existing role indefinitely.

Your role is not guaranteed. Your dignity is.

Mine too.

AI has evolved so damn fast that we’d better ask ourselves:

Where do we actually add value?

For the customer. For the people doing the work. For Grandma Gretchen, still waiting for somebody to fix her heating.

I used to say this in the API world.

I’m saying it again.

START TO ADD VALUE!

P.S. This is not an easy one. We all think we’re adding value. Me included.

But we need to rethink our pretzels.