Like most people who work for a living, I assumed AI was going to be a productivity tool.
You’d write your emails faster. You’d draft your reports in half the time. You’d build the slide deck in fifteen minutes instead of three hours. The framing was always “do more, faster,” and I built it into how I thought about my own week. More output. More throughput. Same brain, bigger lever.
What I’ve actually been seeing is something different. The tools work. They write the email. They draft the report. They build the deck. But the surprising thing isn’t how good they are at it. The surprising thing is how easily they do it, and what that easiness has started to reveal.
A lot of the work was never that hard. It was just laborious.
What the model can do, and what that means
The kinds of work AI does best are interesting if you sit with them for a minute.
Drafting an email that’s mostly polite acknowledgment. Summarizing a meeting where not much was decided. Writing the introduction and conclusion of a report someone needs to file. Producing a slide deck with the standard sections in the standard order. Generating the briefing document that everyone will read for thirty seconds and then forget.
The model can do these things in seconds because they were always patternable.
The structure was conventional. The vocabulary was predictable. The actual content was usually filler dressed up to look like a deliverable. None of this is a criticism of the people who produced these things for years. They were doing what was asked of them, and most of them did it competently. But the work was the kind of work that the model can replicate because it never required much that was unique to the person doing it.
If a sufficiently average prediction machine can produce your output in thirty seconds, the output was never really the thing.
What this is exposing about modern work
I’ve started to think AI isn’t a productivity revolution. It’s an x-ray.
It’s showing what’s actually there. Underneath the busyness, the meetings, the emails, the calendar invites, the documents being passed back and forth, there’s a real question about how much of any given workday was meaningful in the first place. A lot of what looked like output, in retrospect, was the manufacturing of evidence that work was being done.
You produced the deck so people could see you’d thought about it.
You wrote the recap so the meeting felt like it had a deliverable.
You drafted the strategy doc so the team had something to point to.
Some of this had real value. A lot of it didn’t. It was the wallpaper of professional life. It made offices feel productive without anyone needing to ask whether the underlying activity was useful.
When a tool comes along that can produce wallpaper at scale, the wallpaper stops feeling like an achievement. You realise it was always wallpaper.
Why this is uncomfortable
The hard part is that a lot of people built careers, and identities, around producing this kind of work.
It wasn’t just the salary. It was the feeling of being someone who could turn out the report, the deck, the brief, the analysis. There was craft in it. There was effort. You could feel competent at the end of the week, even if nothing the company did was very different because of what you’d produced.
That feeling is harder to hold onto now.
I’ve watched friends in corporate roles try to make sense of this. They’re not stupid. They can see what’s happening. The work they’re proud of getting good at is suddenly something a model can do in less time than it takes them to open a new document. And the question that comes after that, the one nobody wants to sit with, is whether the work was ever as important as it felt.
Most of us, I think, have been quietly afraid of this question for a while.
AI has just made it harder to avoid.
What’s left when the filler comes out
The good news, if you want to look for it, is that something real does seem to remain.
There’s the work of knowing what question to ask in the first place. The model is brilliant at producing answers. It’s not very interesting at deciding what’s worth asking. That’s still a human task, and it turns out to be most of what good work was always about.
There’s taste. Knowing which of three drafts is the right one. Knowing what doesn’t belong in the deck even though the model put it there. Knowing when a piece of writing has feeling in it and when it doesn’t.
There’s judgment. The decisions a model can suggest but can’t make. What to do with the analysis. Whether to ship the product. Whether to have the difficult conversation. Whether the strategy is honest or wishful.
And there’s the actual contact part of the job. The real conversation with a colleague who’s struggling. The honest conversation with a customer. The hard conversation with a partner. None of this scales. None of it benefits from a faster keyboard.
What’s left is smaller, slower, and harder to measure. It also looks much closer to what most of us would call meaningful work.
What I’m starting to think about all this
So here’s the question I’d actually sit with, if I were you. Not whether AI is going to take your job. That’s the easy question, and it lets you off the hook because the answer feels like something happening to you rather than something you have to face.
The harder question is this. What part of what you do all day would you still be proud of if a machine could produce it in the time it takes you to make coffee? Not what you’d defend to your manager. What you’d defend to yourself.
Because honestly, that’s the part nobody is going to answer for you. The model can write your report. It can’t tell you whether the report was ever worth writing. That’s still on you. It was always on you. AI just made it harder to keep pretending otherwise.
