Why you are having the wrong conversation about AI and your workforce.

AI has lost its wonder. Good. Now what?

A friend of mine recently told me a story. He was sitting across from a client, helping her write a cover letter, when he opened a chat window and typed an instruction into one of the more popular machines. He told it to scan the news about the company she was applying to, then sweep through a year of LinkedIn chatter about that company, then crawl Glassdoor for what employees were grumbling about, then pattern-match the lot and tell him what was actually going on under the hood. Thirty seconds later, an answer arrived—competently, even gracefully, proposing the company’s problem with the kind of confidence one used to expect from a consultant on a six-figure retainer.

My friend didn’t blink.

The client did.

She turned to him and asked the question that, six months ago, would have been how did you do that, and is now, increasingly, the only question that matters: how did you know to do that?

The Wonder Is Gone, And Good Riddance

A year ago, someone could build a working prototype of something useful in an afternoon, and the room would gasp. Now they post it on the social network of their choice, and a thousand other men and women who built the same thing yesterday scroll past it on their way to lunch. AI has made it so that so many amazing things require new (less enthusiastic) adjectives. They are now simply available, like a calculator, a spreadsheet, or electricity.

And now AI, itself, is becoming one of these things.

This, I want to argue, is very good news. Not for the people selling the wonder, but for the rest of us.

The Man Who Fired The Wrong Person

A different friend I know, a CEO of a very successful business, recently let go of one of his contractors and, in the cheerful telling of the story afterward, replaced her with an AI agent. He had given the agent a name. He had given it tasks. He had, by his own account, given it the entire scope of what the contractor had been doing, and the result, he said, was that he was saving tens of thousands of dollars a year and producing equivalent or better work.

I argued with him about this.

I told him I thought he had missed the point so completely that he should pay her fees to have the point explained to him again.

He asked me what I meant.

What I meant was this.

He had not hired the contractor because she was someone capable of pushing buttons. He had hired her because she was someone capable of knowing which buttons were worth pushing and applying judgment to the question of whether the result, once produced, was any good.

I think that’s worth repeating:

He had not hired the contractor because she was someone capable of pushing buttons. He had hired her because she was someone capable of knowing which buttons were worth pushing and applying judgment to the question of whether the result, once produced, was any good.

Think about it:

If she were only doing work the agent could do, her value was attributed to the wrong work, and that was a failure of management long before it was a victory of automation. The agent did not replace a person. The agent revealed that the person’s worth was associated with replaceable work all along, which is a rather different and rather more uncomfortable diagnosis.

The Principle, Plainly Stated

The machine does not know what to do. The machine is an extraordinarily competent instrument for executing instructions given by someone who knows what to do. When someone is good, the machine is a force multiplier of an almost embarrassing kind. When someone is mediocre, the machine produces mediocrity at scale, very quickly, albeit with admirable formatting. There is no version of this technology in which the human element disappears. There is only a version in which the human element becomes more concentrated, more visible, and, because there is now far less drudgery to hide behind, more cruelly exposed.

The clients who do well with our software are not the ones who use it the most. They are the ones who knew, before the software arrived, what they were trying to accomplish. The software made them faster or empowered them. It did not make them right.

The Audit, In Five Uncomfortable Steps

None of this is much help to you when you are looking at your team and trying to decide who, exactly, is doing what.

So let me try to be useful for a moment, against my better instincts, and describe how I have started running an audit on my team. Team members: keep reading, the results are good.

It is not elegant. But I have not yet found a substitute.

Take any role on your team and a representative week of that role’s actual output. Not the job description. The output, the documents produced, the messages sent, the decisions logged, and the meetings sat through. Then walk through it the following way.

First. For each piece of work, ask whether the person was executing a known instruction or deciding what the instruction should be. A request that arrives as “do X” is an execution. A situation that arrives as “something is wrong here, figure out what and propose what to do about it” is judgment. Most jobs are a mix. The mix is what you are trying to measure.

Second. For the execution work, ask honestly: this is the part that requires you to stop flattering yourself and your team, and to ask whether a competent operator with a current model and training could produce work of the same quality. Not better. The same. If the answer is yes, that piece of work belongs to the machine, and the only remaining question is whether the person doing it now has the taste to supervise the machine doing it instead. Some will. Some will not. You should know which is which before you make any large decisions, including the decision to do nothing.

Third. For the judgment work, ask the inverse question. What is the person actually doing when they exercise judgment? They are noticing something, a pattern, an inconsistency, a tone, a risk. They are weighing it against the context that the machine lacks. They are deciding what is worth raising and what is worth ignoring. Write down, in plain language, what kinds of noticing this person does that nobody else on your team does, and that no model in your stack can replicate. If you can write a long list, you have an expensive person, and you should probably pay them more. If the list is short or empty, you have a problem, and it’s not the machine.

Fourth. Look at the ratio. Most knowledge roles, in my experience, run somewhere between 60 and 80 percent execution work and 20 to 40 percent judgment work, and the people in those roles spend the bulk of their week on the execution side because that is the work that visibly looks like work.

The single most useful thing you can do, in most cases, is not to fire anyone.

Rather, it is to give the execution work to the machine and to make explicit, in the role itself, that the person’s job is now the judgment piece, the noticing, the deciding, the supervising of the machine’s output. Most people will rise to this. Some will not. The ones who do not were, I am sorry to say, the ones who were always going to struggle in the new arrangement, and you would have discovered this eventually anyway.

Fifth, and last. Run the same audit on yourself. This is the part most leaders skip and the part that matters most. If your own week is sixty percent execution work that a machine could now do, you are not exempt from the diagnosis simply because you sign the checks. You are, in fact, the most expensive button-pusher on your own org chart, and you should be the first person to move. (This is now a sticky note message to myself.)

What’s Actually At Stake

I am told by the more excitable corners of my industry that none of this matters because, soon enough, the machine will know which buttons to push as well.

Perhaps.

I have lived long enough (to the dismay of my knees) to be skeptical of any sentence that begins with soon enough, and long enough also to notice that the people most confident on this point are almost always selling something.

What I can tell you is what I see in my own business, this week, with my own clients. The ones who have stopped wondering at the machine and started thinking harder about the work are pulling away. The ones who have outsourced the thinking to the thing they were supposed to be thinking with are quietly, and I think permanently, falling behind.

The buttons have lost their magic.

Good.

Now we can finally talk about the people pushing them, starting, if you have the stomach for it, with the one in the mirror…

—Ashley Heron
Managing Director, CommaEight