Journal · The AI Architect · 2026-07-12
A wrong first draft is the middle of the work, not the end. Give one clear note, change one thing, and steer.
A wrong first draft is the middle of the work, not the end. Give one clear note, change one thing, and steer.
The tension is not whether the draft is bad. It usually is, and everyone half-expects that. The tension is what happens in the ninety seconds after you read it. Something in you wants a verdict — keep or scrap, good writer or bad tool, worth continuing or not. That verdict-seeking is the whole problem. A first draft was never built to be judged. It was built to be moved.
Stop grading, start locating
When a draft disappoints you, the instinct is to scan the whole thing and form an opinion: this is close, or this is nowhere near right. Both reactions are traps. “Close” makes you tinker at the edges without naming what’s actually wrong. “Nowhere near right” makes you throw the whole thing out and start over, which feels like progress but usually just delays the same problem to draft two.
Instead, treat the draft as a diagnostic. Somewhere in it is the exact spot where it goes off the rails — a paragraph that answers the wrong question, a tone that’s too formal, an example that doesn’t fit the reader you have in mind. Your job is not to grade the draft. It’s to find that spot.
Try this: Read the draft once, all the way through, without editing. Then ask yourself one question: “Where did I first think, that’s not it?” Not where you noticed something wrong on a second read — the first moment, the first sentence or idea that made you flinch. Mark that spot. That’s your note. Not five notes. One.
Change one thing, not everything
Once you have the note, the next mistake is trying to fix everything you noticed while you were at it. You saw the weak opening, sure, but you also noticed the second paragraph rambles, and the ending is flat, and maybe the whole structure should be different. Resist this. If you change five things at once, you can’t tell which change helped and which one hurt. You’ve just generated a new draft with a new, more tangled set of problems.
Pick the one note. Make the one change. Then look at the result before deciding what’s next. This is slower in the moment and much faster overall, because you’re not guessing anymore — you’re testing.
This applies whether you’re revising your own writing or working with an AI tool that generated the draft. If you’re prompting an assistant, resist the urge to write a paragraph of feedback covering tone, structure, length, and word choice all at once. Give it the one note: “The opening buries the point — start with the point.” Let it revise. Look again. Give the next note. This is not slower than a big list of feedback. It’s usually faster, because each round is clean enough to actually evaluate.
Try this: Next time you get a draft back — from yourself, a collaborator, or a tool — write your feedback, then cross out every sentence except the first one. Send only that. See what comes back. You’ll often find the other four notes were either premature or would have solved themselves once the first change landed.
Steering is a direction, not a destination
The word “steer” matters more than it looks. Steering assumes movement already exists — you’re not starting the car, you’re adjusting where it’s headed. This reframes what a draft is for. A draft’s job is to exist, badly if necessary, so that you have something to push against. You cannot steer a blank page. You can only steer a thing that’s already moving in some direction, even the wrong one.
This means the wrongness of a first draft is not a failure of the process. It’s the process working. The draft’s rough shape tells you what the shape isn’t, which is often the fastest way to find out what it is. People who wait for a strong first draft before they’ll engage with it are waiting for a car that’s already pointed the right way before they’ll agree to drive. That’s not how steering works.
Try this: Before you start a piece of writing, tell yourself explicitly: “This draft’s job is to be wrong in an informative way.” It sounds like a small mental trick, but it changes how you read your own early work. You stop bracing for disappointment and start looking for the information the wrongness contains.
None of this makes revision effortless. It still takes attention, and some days the one true note is hard to find. But the frame changes what the work feels like — less like judgment, more like navigation. The book goes further into how this plays out across longer projects, where the notes compound and the steering has to happen many times in a row, but the core of it starts here: find the spot, make the one change, look again.
Go deeper. The full method is in The AI Architect. New here? Start with the free companion pack, or explore the series.