What I Actually Do With AI All Day

My friend from a non-tech background asked me last week, half-joking, what I do for eight hours when AI does the work.

He was joking. But also not really. The question lives in everyone’s head right now, including mine. The honest answer is more interesting than either of us expected. So I told him I would write it down.

This is that answer.

The work shape changed. The hours did not.

Most of what I now do all day is something I would not have called work three years ago.

Here is what a typical Monday looks like.

I open my queue. There are about twenty open items, most of them flagged, sorted, and pre-summarised by an AI system. Three years ago this kind of intake would have been a person’s whole job. Now it is the first hour of mine. But I am not reading the queue for the answer. I am reading it for what the AI got wrong.

That sentence is the whole essay. Let me write it again.

I am not reading for the answer. I am reading for what the AI got wrong.

This is the work now. Not producing the first draft. Catching what the system cannot catch. The thing that looks correct but does not account for the customer’s actual context. The case that pattern-matches to a known issue but is something subtly new. The empathy gap. The judgment call.

What disappeared was the bottom of the ladder

Six years ago, you would ladder up the seniority chain to access this kind of judgment. Junior people did the easy ones. Senior people did the hard ones.

AI now does the easy ones. So the entry-level role disappears. And the senior judgment has to expand to fill the day.

This is what the data on entry-level hiring is showing. Not because seniors got better. Because the bottom of the ladder evaporated.

There is a frightening implication in this that nobody talks about clearly. Senior judgment is not made out of thin air. It is built from years of doing the entry-level work. If the entry-level work disappears, where do tomorrow’s seniors come from? I do not have an answer to that. I do not think anyone does.

But it is the question the next decade of work hinges on, and very few people are asking it out loud.

Taste is the new entry-level

The other half of my day is writing. Documentation, internal updates, customer-facing communication.

Two years ago this was the slow part of the work. Now I draft six versions of an update in the time it used to take to draft one. But the version I publish is one I put most work into, because the other versions generated by AI are fine, and “fine” does not meet my standards.

Fine does not compound. Taste does.

This is the part the AI hype gets wrong, and the AI fear gets wrong too. The hype says: the AI will do everything for you. The fear says: the AI will replace you. The truth is more boring. The AI will do the work that any competent person could do. What you bring is judgment that took years to develop and shows up in three sentences of edited output.

For now, that is enough. I do not know how long that lasts.

The skill is not using AI. It is knowing what good output looks like before you see any output. The AI gives you a hundred drafts. Taste tells you which one is the right one. People who have the AI but not the taste produce a lot of mediocre work very fast. People who have the taste but not the AI produce excellent work too slowly.

The combination is the new shape of expertise.

A real workflow, since people ask

Morning queue. AI summarises twenty items in three minutes. I read all twenty summaries in fifteen. I correct the misreads. The misreads are usually the ones where the customer is angry but technically incorrect, and the resolution requires saying so without losing them. AI does not do that part well yet.

Afternoon writing. Draft a memo. AI produces a structure. I keep the structure, throw out the language, write it in my voice. Total time: forty minutes for what used to take ninety.

Evening, my own work. I write a Substack essay. AI organises my journal notes into rough sections. I write the sentences. The AI does not write the sentences. If you can read this essay and tell which parts are mine, all of them are mine.

What I tell my friend now

The era of one person doing the work of a team is not coming. It is here. It just looks like one tired person at a desk on a Monday morning, drinking coffee, deleting most of what the machine produced.

The work has not disappeared. It has been redistributed.

Most of it now lives in the part you cannot outsource. Which is, I suppose, the part it should have lived in all along. The mechanical execution was always the lowest-value part of the job. We just could not see it clearly because the mechanical execution was so loud. Now the mechanical execution is quiet, and the only thing left in the room is the part that always mattered: judgment.

The next time my friend asks what I do all day, I will tell him I make decisions. Hundreds of them. About what the AI got right, what it got wrong, what to publish, what to throw away, what is worth a customer’s time and what is not. The decisions used to be invisible because the execution was so visible. Now the decisions are the only thing left.

This may turn out to be a temporary configuration. AI may eventually take the decisions too. I am preparing for that. But preparing for it does not look like panic. It looks like building something else, slowly, in the hours after the day job ends.

The work moved.

So did I.