I mean, the reasons were in there somewhere. I just couldn't get them out in any order that held together, and after enough conversations I started to suspect I didn't actually know what I thought.
Then I sat down to write about it, and by the time I was finished I had a position.
Not a tidy summary of what I'd always believed, an actual argument I now hold and can defend when someone pushes on it.
The piece is where the conclusion got made, rather than a place I filed one I already had.
What I Already Believed About Writing Things Down
I've made the case twice before that builders should write more.
Once for capturing the mess while you're still inside it, because the version you write six months later has been sanded down by your own ego into something useless.
And then again for logging your reasoning before you ship, so that when a feature flops you can trace it back to the decision that caused it instead of guessing.
Both of those are true, and I still do both. But look at what they assume.
They treat writing as storage. A container you pour finished thoughts into so you can retrieve them later. The decision log template I published has fields for your hypothesis, your alternatives, your key assumptions, and the whole design implies you walk up to that document already knowing those things.
That's not how it goes.
Half of what ends up in those boxes only exists because I sat down to fill them in. The composition researchers figured this out decades ago, describing writing as an active process of meaning-making rather than the transcription of pre-formed thoughts, where ideas get generated and revised in real time as the sentences take shape.
The Sentence Makes You Commit
A thought in your head can stay comfortably vague forever.
It can be a feeling with an argument sort of implied around it. A direction rather than a claim.
You can turn it over for months and never notice the gap in the middle, because the shape of it feels complete from the inside.
Writing takes that away.
The moment you commit a thought to a sentence you have to pick a verb, decide what causes what, and finish the clause. The vagueness has nowhere left to hide.
Paul Graham put the same idea in blunter terms when he noted that half the ideas that end up in an essay are ones you thought of while writing it, which is a strange thing to admit until you've watched it happen to you.
Most of the time, that's when you find out the thought doesn't hold.
You write your reason down, look at it sitting there in plain language, and realize you've produced the socially acceptable version rather than the true one. The line you'd been repeating because it sounded fine out loud.
So you write the next one. That fails too, but it fails differently, and the way it fails tells you something about what you're circling.
Watching your own explanations collapse on the page is the thinking.
There's no version where you skip to the good answer without producing several bad ones first.
With the day job piece, the early attempts were all defensive.
They truly read like a man explaining why he hadn't done the thing he was supposed to do. It took several rounds of that before I noticed the defensiveness itself was the tell.
I'd been apologizing for a compromise when what I actually had was a configuration I had chosen on purpose and would likely choose again. The argument only showed up once I'd written the weak version enough times to see what was wrong with it.
Where AI Actually Sits In This
I want to be specific about this part 'cause the general version of this conversation is useless.
My process looks like this.
I brain-dump first, and the dump is truly raw. Half-formed reasons, contradictions I haven't noticed, three different framings of the same idea, notes that trail off, etc. Then I hand that pile over to AI for analysis and ask what patterns are in it, what I keep circling. I'm essentially looking for what the actual argument underneath might be.
What comes back is a read on my own mess, sometimes with a rough shape for how it might be organized.
Then I write the thing. All of it, sentence by sentence, in my own words, and that stage is where most of the work happens.
The raw material comes out of my head before any model touches it. Nobody can generate my particular confusion about my own life, and the confusion is the input the process runs on.
What comes back is analysis performed on material I produced, which makes it a mirror rather than a ghostwriter.
Sometimes the analysis names a pattern I'd been circling for weeks without seeing. That connection came from outside my head.
I produced the material and I'll do the writing, but I didn't spot the thread. Calling that entirely my own thought would be a lie, and calling it entirely the machine's would ignore that it only existed because I'd spent weeks generating the mess it was found in.
The honest answer is that it's collaborative in a way that doesn't have a clean precedent, and I'd rather sit in that than pretend otherwise. This stuff is pretty new. To me, to you, to mostly everyone if we're being honest here.
What I'm confident about is where it stops being collaborative.
A named pattern arrives as a candidate rather than a position and it becomes something I believe only after I've written it out in full and either accepted it or fought it, and the fighting is where most of my actual conclusions get made.
The Version That Breaks
There's a much shorter version of this that produces nothing at all.
Skip the dump. Skip the writing.
Describe the topic in a couple of sentences, ask for a finished draft, tidy the result, publish it. The output looks roughly the same from the outside. Nobody reading it can really tell what happened upstream.
The difference is entirely internal. You have an artifact and you never had the thought.
Nothing moved in your head, no position formed, and the next time someone asks you a hard question about your own argument, there's nothing underneath it to reach for.
I keep coming back to the fact that this is the exact same tool in both cases.
The model doesn't change. What changes is where it sits in the process.
Point it at material you generated and it's a mirror that shows you patterns in your own mess. Point it at an empty page and ask for the mess itself, and you've handed over the part that was doing the work.
Scientific publishers have started asking the same question about research papers, wondering whether a manuscript written by a model means readers are absorbing the machine's thoughts instead of the researcher's. The stakes are lower on a personal blog, but the mechanism is identical.
The typing was never the expensive part. It's just the most visible one, which makes it the easiest thing to try to save.
The Fight With Your Own Draft
The most underrated stage is what happens when you read back something you wrote yesterday and can't fucking stand it. Yeah, it happens more than you'd think.
I kill openings constantly. I cut paragraphs I wrote that sound too generic or like nobody in particular wrote them. I read a framing I committed to an hour earlier and realize I don't believe it anymore.
That resistance feels like editing, or like fussing over prose, and for a long time I thought that's all it was.
But that resistance is where my positions harden.
You can't discover that you disagree with something until it's sitting in front of you in finished sentences.
A vague notion gives you nothing specific enough to push against. A completed paragraph gives you something to reject, and the rejection is informative.
Every time I've thrown out an opening I wrote because by the end of it it no longed matched my actual experience, the thing I wrote to replace it was closer to what I really thought.
This is also where outside feedback earns its place. Once a draft exists, I'll hand it over and ask what's weak, what I'm hedging on. I want to know if my arguments thin out somewhere.
Sometimes I take the note. More often the value is in disagreeing with it, because defending a paragraph is how I find out whether I truly meant it.
That's why the finished piece so often argues something the outline didn't. The outline was a hypothesis, and writing it out is the experiment that tests it.
Back to the Question
I hold a position on the day job now that I didn't hold three years ago, and I can point at exactly where it got made.
It didn't happen in the shower, and it didn't happen in any of the dozens of conversations where I gave someone a shrug with sentences attached.
It happened in a draft, over several passes, while I watched my own weaker explanations fall apart and kept writing until something held.
So if you're carrying a question you can't answer cleanly, I'd stop waiting for the answer to arrive on its own. Turning it over in your head is exactly the condition that lets it stay vague, because nothing in there is forcing you to finish the sentence.
The hardest part is usually getting past the activation energy of starting, which is why the advice to write what you don't know works better than it should.
Sit down and write the bad version. Then write the next one.