What Happens After AI Does the Work
There is a strange moment that happens almost every time I use AI now. I ask it to research something, structure an argument, draft a presentation, analyze a market, or turn a pile of half-formed thoughts into something coherent. Thirty seconds later, I am staring at an output that would have taken me hours a few years ago. It is organized. It is articulate. It looks suspiciously finished. And for about twelve seconds, I feel like I have discovered some kind of productivity cheat code.
Then I start working on it. The facts need checking. One assumption feels too convenient. Three paragraphs say essentially the same thing with increasingly impressive vocabulary. The recommendation sounds reasonable until I imagine saying it to someone who actually has to implement it. A sentence is technically accurate but gives me that vague, irritating feeling that something is wrong. So I rewrite it. Then I rewrite the rewrite. Eventually I realize I have spent 25 minutes debating a word that AI generated in less time than it took me to open the document.
AI makes the first 90% easy. Increasingly, the last 10% becomes the whole job.
Creation Is Getting Cheap
For most knowledge work, creating a plausible first version used to consume a meaningful portion of the effort. Someone had to find the information, organize it, stare at a blank page, create the structure, write the first draft and then slowly improve it. AI collapses a lot of that work into minutes. That is an enormous productivity gain, and I don't think we should minimize it. I use AI constantly precisely because it removes work I never particularly enjoyed doing. Starting from something is much easier than starting from nothing. Give me a mediocre draft and I can attack it. Give me a blinking cursor and suddenly checking the refrigerator feels like an important business activity.
The weird consequence is that creation itself becomes less valuable. If everyone can create a competent strategy summary, customer brief, presentation outline, article or competitive analysis in minutes, producing one is no longer particularly impressive. The threshold moves. And I think that is where some of our current excitement about AI productivity gets a little ahead of itself. We measure how quickly something appeared. We rarely measure how long it took before someone knowledgeable was willing to put their name on it.
The Last 10% Is Mostly Judgment
That final stretch contains way more human work than we probably anticipated. You have to know which source is credible, which assumption deserves another look, what context is missing and whether the conclusion actually follows from the evidence. You need enough experience to recognize when something is perfectly logical and still completely stupid in the real world.
There is taste involved too, which is annoyingly difficult to quantify. Knowing what to remove matters. Knowing when a presentation has one idea too many matters. Knowing that a paragraph sounds impressive but says absolutely nothing matters. Increasingly, I find myself using AI to generate more possibilities and then spending my time deciding which ones deserve to survive.
It reminds me a little of owning a robot vacuum. The thing can clean 90% of the floor while I am doing absolutely nothing, which still feels vaguely magical. But I barely notice the part it cleaned anymore. I notice the 10% it missed. The corner it refuses to enter. The shoelace it somehow finds every single time. The chair it has decided is an impenetrable architectural barrier. So my role in vacuuming has changed. I no longer clean the whole floor. I deal almost exclusively with the weird exceptions the machine cannot handle, which makes that tiny remaining piece feel like the entire job.
AI has started to make knowledge work feel strangely similar. Most of the visible progress happens quickly. The remaining pieces require disproportionate thought because those are the pieces where the decisions are hiding.
The Job Is Moving
I am increasingly convinced that AI does not eliminate expertise nearly as cleanly as some people hope. It changes where expertise gets applied. Less time may be spent assembling the raw material, while more attention goes toward interrogation, refinement, prioritization and judgment. That changes what being “good” at knowledge work means. Speed still matters, probably more than ever. But speed to first draft is becoming cheap. The advantage moves toward people who can recognize what deserves to happen after the first draft appears.
AI gets us to 90% astonishingly fast. The competitive question may increasingly be what we do when we get there.
