I used to be proud of opening a blank document.
There was something in it — the cursor blinking, no scaffolding, just me and the problem. That felt like real thinking. AI-generated text, by contrast, felt like cheating. Or worse, like wearing someone else's shoes. The output would be theirs, not mine.
This belief is extremely common among professionals who are smart enough to care about quality. And it is wrong. Not a little wrong — structurally wrong, in a way that costs real time and produces, on average, worse first drafts, not better ones.
Here's the actual mechanism. Thinking is metabolically expensive. When you open a blank document and try to think and write simultaneously, your brain is doing two hard jobs at once: generating the ideas, and shaping them into language. Those tasks compete. The result is usually a draft that is either over-structured (you reach for the template in your head) or under-structured (you free-associate and then spend forty minutes hacking it into something coherent). Neither is "pure" thinking. Both are ways of managing cognitive load.
What AI does, when you use it as a first-move tool rather than a last-resort one, is separate those two jobs. You prompt it roughly — barely a sentence, sometimes just a fragment — and it comes back with something. That something is almost never quite right. But now you are reacting, not generating. And reacting is dramatically cheaper than originating. You can see what's wrong with a sentence in three seconds. Writing the right sentence from nothing can take three minutes.
This is the mode-switch that most people miss. The point of the AI draft is not that it's good. It's that it exists. You can push against it. That friction is productive.
The reason the myth sticks is interesting. It comes down to what psychologists call the endowment effect — we overvalue things we feel we own. Once an idea has come from inside your own head, it feels more like you. A thought you generated is worth more to you than the same thought handed to you, even if they're identical. So the blank-page habit gets reinforced: your draft, however painful to produce, feels like genuine intellectual property. The AI draft feels borrowed.
But ownership of an idea doesn't transfer at the point of generation. It transfers at the point of decision. When you read an AI draft and say "not this, but something like this, except—" — that's where your thinking lives. That's the part that's actually yours. The generation was just scaffolding.
I want to be honest about the complication here, because there is one. Some tasks really do benefit from starting cold. If you're trying to generate genuinely novel combinations — a research hypothesis, a creative brief that can't resemble anything that came before — then feeding AI your problem first can flatten the possibility space before you've had a chance to roam it. The model pulls toward the probable. Sometimes you need the improbable.
So the corrected belief isn't "always use AI first." It's "the blank page is not neutral." It has costs. Those costs are worth paying sometimes, and a habit of paying them by default is just friction wearing the costume of rigour.
The working test I use: if I'm going to need to communicate this eventually — a document, an email, a recommendation, a plan — start with a rough AI draft. If I'm trying to discover something I don't know yet, or produce an idea with no prior art, start alone. Most professional tasks are the first kind. Most of us pretend they're the second.
The broader shift underneath this is moving from AI as a polishing tool to AI as a thinking partner at the start — not the end. That sounds small. It changes your daily output significantly.
It also takes practice, because you have to get comfortable with drafts that are clearly not you yet. That's the skill: knowing what to reject, what to keep, and what the gap between the two tells you about what you actually think. The model is not doing your thinking. It's giving your thinking something to work against.
If you want to build that habit in a way that actually sticks, that's precisely what The AI Habit is designed for — 15 minutes a day, working through real skills in real contexts, with the first 14 days free. No theory for theory's sake. Just practice that compounds.
You can start here: https://harrysharman.com/projects/ai-habit/