There is a way of using AI that feels productive and is actually making you a little worse at your job every single week. You already know the one. You open a chat window, you type a prompt, the machine hands you a clean draft in eight seconds, you skim it, you nod, you ship it. Felt fast. Felt like leverage. Then a month later you reread what you sent and you can barely find yourself in it. It was fine. It was competent. It was completely forgettable. And somewhere along the way the muscle you used to use to think the thing through started to atrophy, because you handed the thinking to a tool that was never built to do your thinking for you.
I want to talk about the other way to use it. The way that actually compounds. It is the single highest leverage shift I have made in the last year, and it costs you nothing but the willingness to do the hard part first.
Here is the whole thing in one line. I create. AI critiques. I refine. Three moves, in that exact order, and the order is not a suggestion. The order is the entire game.
Why you create first
When you let AI write the first draft, you outsource the most valuable thing you own, which is your judgment about what matters. The machine does not know which detail will land with your specific client. It does not know the story you told on the sales call that made them lean in. It does not know the thing you are secretly avoiding saying because saying it out loud makes the project real. It averages. It gives you the middle of the road, smoothed and safe, because that is what it was trained to do. The middle of the road is exactly where nobody remembers you.
So you go first. You write the rough version. The ugly one. The one with the half-formed sentence and the point you are not sure how to make yet. That rough version is where your actual thinking lives, and your thinking is the asset. It does not need to be good. It needs to be yours. You are not trying to produce a finished product on the first pass. You are trying to get the raw material out of your head and onto the page so there is something real to work with.
This part feels slower. It is slower, in the moment. But you are not optimizing for the next eight seconds. You are optimizing for work that sounds like a human being made it on purpose. That is the whole point.
There is also a voice problem nobody warns you about. Every time you let the machine draft for you, a little of your own voice gets sanded off, because the model writes in the smoothed-out average of everyone it ever read. Do that enough times and you wake up one day sounding like a competent stranger. Your voice is not a luxury. In a world where anyone can generate infinite competent content in seconds, your specific way of seeing things is the only moat you have left. Creating first is how you protect it. You cannot lose a voice you used to write the first draft.
What “critique” actually means
Once the rough draft exists, now the tool earns its keep. And the way most people get this step wrong is they ask AI to “make it better,” which is a useless instruction that produces useless results. “Better” just means “more average,” and average is what you are trying to escape.
Instead, you point it at specific weaknesses. You make it work like the sharpest editor you have ever had, the one who does not care about your feelings and only cares about whether the thing holds up. Here are the prompts I actually run:
“Where is this unclear? Point to the exact sentences a busy reader would have to read twice.”
“What am I assuming the reader already knows that they probably don’t?”
“Where am I hedging or being vague because I haven’t actually decided what I think?”
“What is the strongest objection to my main point, and have I addressed it?”
“What did I leave out that the reader will be wondering about?”
Notice that none of those ask the machine to rewrite anything. They ask it to find the holes. A good critique does not hand you a new draft. It hands you a map of everything you were too close to see. The blind spot is the deliverable. You are using a tireless, unflappable second set of eyes to catch the stuff your own brain quietly skipped over because you have read your own sentence forty times and your eyes glaze.
This is also why it works at any hour, on any project, without you having to feel guilty about asking a human colleague to review your fourth revision of a cold email. The tool does not get tired and it does not get annoyed. Lean on that. Tools like Claude and the rest of the current crop are genuinely good at this kind of structured critique when you ask them precise questions instead of lazy ones.
Why you have to refine it yourself
Here is the trap on the back end. The critique comes back, it is sharp, it found four real problems, and now there is a powerful temptation to just say “okay, fix all of that” and let the machine rewrite the whole thing. Do not do that. The second you do, you are right back to shipping the average version, except now it is dressed up in the confidence of having been “edited.”
You take the critique and you make the changes. By hand. In your own words. Because the point of the critique was never to get a list of fixes. The point was to see your own work more clearly so you could make it sharper in your own voice. When you fix it yourself, two things happen. The work stays yours, and you get a little bit better as a writer and a thinker, because you just practiced the exact skill the tool was tempting you to skip. Outsource the spotting. Never outsource the deciding.
What this looks like on a real Tuesday
Say I am writing a proposal for a consulting client. Old way, I would have typed “write a proposal for X” and gotten a generic template with my client’s name pasted into the blanks. New way, I write the rough proposal myself, scoped to exactly what we talked through on the call, including the awkward part where I have to tell them their current process is the thing holding them back. It is messy. It rambles in two spots.
Then I run it through the critique prompts. The feedback tells me my pricing section assumes they already understand the ROI math, which they don’t, and that I buried the single most important sentence in the third paragraph where nobody will see it. Both true. Both things I could not see because I wrote it. So I rewrite the pricing section to walk them through the math, and I pull that buried sentence up to the top where it belongs. Twenty minutes. The proposal that goes out is unmistakably mine, scoped to them, and a level sharper than anything I would have shipped on the first pass or anything the machine would have handed me cold.
Same loop works on a sales email, a process document, a podcast outline, a hard message to a team member. Same three moves every time. Create, critique, refine. The raw material comes from you. The second set of eyes comes from the tool. The final call always comes back to you.
And it works on the stuff that is not writing at all. I run my own decisions through it now. I will dump my messy reasoning about whether to take on a client or restructure an offer, and instead of asking the machine what I should do, I ask it to poke holes in my logic. Where am I being optimistic? What am I not accounting for? What would a skeptic say? It is the cheapest board of advisors you will ever have, and it never once tells you what you want to hear if you ask it the right way. The skill is in the asking. Lazy questions get lazy answers. Sharp questions get a mirror.
The part nobody wants to hear
This workflow is slower than copy and paste. That is not a bug. The whole reason it produces better work is that it keeps you in the chair doing the thing that is actually hard, which is thinking clearly and then saying what you mean. AI did not remove that work. It just made it really easy to avoid, and a lot of people are quietly choosing to avoid it and calling the avoidance “efficiency.”
If you want a clean test for whether you are using these tools well, ask yourself one question after you finish a piece of work. Did the tool help me say what I think more clearly, or did it think for me? If it is the first one, you are building a skill and an asset at the same time. If it is the second one, you are renting competence you do not own, and the bill comes due the day you have to do the thing without the machine in the room.
Capture your raw thinking somewhere it does not get lost, whether that is a notebook, a voice memo, or a tool like Fathom pulling the real language straight off your calls. Then run the loop. You create. The machine critiques. You refine. Protect that order and you will spend this year getting sharper while everyone around you quietly gets more generic.
That is the tactic for the week. Not a new tool. A new sequence. Go run something through it before you ship it today, and notice how different the finished thing feels when your fingerprints are still all over it.
One step, one day. Grace over guilt.
— Dan Kaufman
Tools I actually use and trust. If you sign up through these links I may earn a small commission at no extra cost to you.