Writing · AI and human responsibility
Everyone Is Becoming a Manager, or a Puppet
By Richard K. Marshall · Originally published
You’re either the manager or the puppet.
I keep coming back to something I think we’re all going to have to figure out.
AI is doing more and more of the work.
Not just answering questions. Actually doing things. Writing the email. Researching the company. Filling out the form. Scheduling the appointment. Building the thing. Maybe eventually buying the thing.
And that changes your job.
Because somebody—or something—can now do work that used to require you personally sitting there doing it.
But here’s the part I don’t think we’ve fully absorbed yet:
You still own the outcome.
So I think we’re all drifting toward one of two roles.
You’re either the manager.
Or you’re the puppet.
The puppet lets the system run the show.
The AI drafted it, so they sent it.
The system approved it.
The dashboard is green.
The agent said it was done.
And somewhere along the way, nobody can really answer the most important question:
Who’s responsible if this goes wrong?
“The AI did it” is not going to be a very useful answer.
The manager is different.
The manager lets the tools do the work they’re good at, but keeps hold of judgment and responsibility.
AI can research something for me.
It can draft something for me.
It can recommend something.
It can probably do a lot of things I used to have to do myself.
But somewhere, there still needs to be a human being who says:
Yes. Do that.
And if it turns out to be wrong:
That’s on me.
That’s really the whole thing.
Delegation is not abdication
There’s a big difference between saying:
“I had AI help me do this.”
and saying:
“The AI handled it.”
One of those statements still has a human being in charge.
The other one is somebody trying to hand responsibility to a machine.
You can delegate work.
You can automate work.
You can give a system permission to do certain things.
But I don’t think you can automate responsibility.
That still belongs to somebody.
My own rule is the Marshall Principle:
Artificial intelligence may assist human decision-making, but responsibility always remains with humans. Authority cannot be automated.
That sounds obvious.
I don’t think we’re acting like it’s obvious.
“Done” is doing a lot of work these days
Another thing that worries me is how quickly systems declare victory.
Something says:
Completed.
Green checkmark.
Everybody moves on.
But “completed” can mean several very different things.
A human intended for something to happen.
The system attempted to do it.
The system says it completed the job.
And then there’s the question nobody wants to ask:
Did the thing actually happen?
Those aren’t the same thing.
An email can be sent.
That doesn’t mean it was received.
It can be received.
That doesn’t mean it was read.
It can be read.
That doesn’t mean it accomplished anything.
AI systems are going to make this problem worse because they’re extremely good at producing things that look finished.
A beautiful report is not necessarily a good report.
A generated answer is not necessarily true.
A completed task is not necessarily a completed outcome.
Assertions are not evidence.
That one matters.
The question I think we need to start asking
Before you turn an AI loose on something consequential, ask:
Who owns this?
Not who built the system.
Not what model generated the answer.
Not which agent ran the workflow.
Who owns the outcome?
Put a human name on it.
Then ask another question:
Can the system do this—or has somebody actually authorized it to do this?
Those are different things too.
Just because a tool can send the email doesn’t mean it should be allowed to send the email.
Just because an agent can spend the money doesn’t mean you’ve decided it may spend the money.
Capability is not authority.
I suspect we’re going to learn that lesson repeatedly over the next few years.
Let the machines move faster. Keep the humans accountable.
I’m not arguing that humans should sit around approving every mouse click.
That would be stupid.
The whole point of these tools is to get useful work off our plates.
Let the system research.
Let it draft.
Let it compare.
Let it watch things.
Let it remind you.
Let it do the repetitive work.
But think carefully about where the line is.
Sending something into the world.
Spending money.
Publishing something under your company’s name.
Deleting something important.
Making a commitment to a customer.
Those are different.
The more irreversible the consequence, the more I want to know who made the decision.
And I want evidence when the work is supposed to be finished.
Not just a green checkmark.
Maybe that’s the real new job
Everybody has been talking about prompt engineering.
Maybe that’s not the important job.
Maybe the important job is learning how to manage intelligence that isn’t human without pretending it somehow became responsible.
Because it didn’t.
The machines are going to get better.
They’re going to become more capable.
They’re going to be able to do things that would have sounded ridiculous a few years ago.
And every time that happens, we’re going to have the same choice.
Do we stay in charge of our judgment?
Or do we slowly become passengers in systems we’re supposedly operating?
I know which one I want to be.
The manager.
Not the puppet.
— Richard K. Marshall Marshall Intelligence · Lexington, Kentucky
More on keeping humans in charge of AI at marshall.net.
Originally published on X: https://x.com/RichMarshall/status/2095578372038443164 · . Refreshed .