Here is the thing nobody tells you: the days don’t get easier. They get stranger.
When you first hear that someone delegates eighty percent of their work to an AI, you picture a person leaning back, feet up, skimming outputs and approving things. You picture leisure dressed up as productivity. And in one narrow sense — the sense where “work” means keystrokes, first drafts, look-ups, reformatting, summarising — the picture is accurate. Those tasks vanish. The hours they occupied do not.
What happens instead is that the cognitive texture of the day changes in a way that is genuinely hard to describe without sounding like you are complaining about something enviable. The best analogy I have found is the director versus the actor. An actor’s job is hard and physical and rhythmically satisfying — you do the thing, you feel it, you go home. A director’s job is hard in a different register: you have to hold the whole film in your head, notice when a performance is two degrees off what you wanted, and then find language precise enough to correct it. The correcting is not the hard part. The noticing is not the hard part. The language is the hard part.
Working heavily with an AI assistant is mostly a language problem. The model is very fast. It is very literal. It is, in my experience, completely undisturbed by being wrong — not because it lacks something, but because it lacks the kind of ego that makes being wrong uncomfortable. You say “do this,” it does something adjacent to this, and it presents the result with the same steady confidence it would bring to a correct answer. Which means the burden of detecting the adjacency falls entirely on you. The bottleneck, which used to live at execution — how fast can I write this, how long will this take — moves upstream into articulation. What do I actually want? How do I say that with enough precision that the gap between my intent and its output closes?
This sounds manageable until you do it for eight hours. Articulation at volume is its own fatigue. You finish the day having produced more than you could have alone and feeling — not drained exactly, but scoured. Like you have been thinking with a very fine instrument all day and your grip has gone soft.
There is also something that happens to your relationship with your own expertise. The AI does not know what matters in the way that experience teaches you what matters. It knows what is documented. So you find yourself supplying the judgment that sits beneath the surface of any domain — the intuitions, the pattern-matches, the “this is technically correct but will land wrong in practice” — and you find that you had more of this than you realised, because before, it was embedded quietly in your execution. When execution is handled externally, the judgment has to become explicit. You have to think about what you know, not just deploy it. That is interesting and occasionally uncomfortable, the way that writing about a belief is more demanding than simply holding it.
Where this leads, I am genuinely uncertain. There is a version of this trajectory where articulation gets easier with practice — where you develop a fluency in specifying intent that becomes its own skill, clean and transferable. There is another version where the distance between thinking and doing keeps widening, and something atrophies quietly in the gap. I do not know yet which version I am in. Some days it feels like the former: I am learning to think in a more precise register, to be explicit about assumptions I used to carry silently. Other days it feels like the latter: I am becoming a better editor of work I did not do, which is not quite the same thing as becoming better at the work.
The strange part is not that AI does more of the work. The strange part is what you learn about yourself when it does — what was skill, what was habit, what was just the satisfying friction of moving through a task. The question I keep coming back to is not “how much of this is sustainable?” It is: what exactly am I practicing, on the days when I am mostly directing? And what, very slowly, am I not?