The invitation had been sitting in my inbox for a week, unopened in the way that the things you mean to do start to accuse you. A webinar for Chiefs of Staff. It was the title that stopped me, because it was a sentence I’d heard before, from a founder, in that slightly impatient tone people use when they’ve decided the future is now your problem: where to start when your boss says, “figure out AI.”
I missed it. School run, a client call that ran long, the usual. By the time I remembered, the tab was one of fourteen I was pretending to have under control.
The phrase stayed, though.
Every Chief of Staff I know has been handed some version of it this year.
Figure out AI.
Two words and a verb doing an enormous amount of unpaid labour. It sounds like a task and behaves more like a mood. The very natural instinct is to reach for tools, because tools are visible and a mood is not, and so you go and find the thirty-seven newsletters and the leaderboard of models and the man on LinkedIn who’s automated his whole life and seems, on balance, less happy for it.
The webinar promised the sensible things, and they are sensible. A time audit, so you can see where your week goes rather than where you think it goes. A list of what to hand off first. A ninety-day plan. A way to tell the story in numbers for the people upstairs who only move when the number moves. I’ve done versions of all of it, I’d recommend it, and none of it’s the answer to the question the founder was really asking. He didn’t know he was asking it. They rarely do.
When I ran my own time audit, the result was almost embarrassing in how obvious it was. The hours were going where the skill wasn’t: status updates, a deck someone else had half-built that I was tidying, the fourth follow-up to a person who was always going to reply on Friday regardless. Hand all of that to a machine and as it turns out, the machine is very good at the parts of my job that drain me and useless at the parts that are the job.
A colleague of mine, who I’ll leave unnamed for now, automated her entire board-pack process, beautifully, and what she said about it afterwards has stuck with me. The pack takes her ten minutes now, she told me, almost annoyed about it. The problem, she said, is that she used to do her thinking while she built it. She was tapping a pen against the desk the way she does when she’s worked it out and doesn’t like the answer. The building was the thinking. Nobody puts that on the ROI slide.
The frameworks leave this part out, and it’s the bit I’d stay up arguing about. Once you’ve automated the things that drain you, you’re left holding the things that don’t, and those turn out to be the whole reason you’re in the room. Someone has to decide which of the model’s three perfectly reasonable options is the one that won’t blow up in six months. The machine produces options but a human carries the decision, and the consequence, and the apology if it’s wrong.
It gets sharper than that, fast. The speaker came from the world of hiring technology, and the framework she was offering ran right across the talent lifecycle, hiring, firing, all of it. I think that’s exactly the right place to look and exactly the place to be most careful, because that’s where figuring out AI stops being about your calendar and starts being about someone’s livelihood. You can let a tool sort a thousand CVs but you cannot let it tell a person, on a Tuesday, that their job is gone – and you shouldn’t want to. The nerve to make that call, and to make it like a human being, is the job underneath the job.
It surfaces in smaller ways, too. Knowing which of the four AI-drafted versions of a message sounds like your CEO and not a press release is taste, and taste is stubbornly, gloriously human, and no audit I’ve ever run goes looking for it.
So, where do you start? You decide, on purpose, what you won’t hand over. The audit tells you what to automate. The harder, stranger question is what you’d protect even if the automation worked perfectly and everyone else handed it over without blinking.
I never did watch the recording. I will, probably, on a Friday, with the half-attention these things tend to get. But I think I already know the part it won’t cover; the part no webinar covers, which is that “figure out AI “was only ever a question about you, and what you’re for, dressed as an instruction about software. Whoever asked it didn’t have the words for that and most of us don’t.
We get handed the question anyway, which is its own kind of trust, and we go looking, and the looking turns out to be the job.
