How these blogs actually get written
The voice you read in these posts is mine. The typing is not. Here's how that actually works, because I think it matters and I keep getting asked.
I rant. Often.
Anyone who knows me knows the pattern. I get an idea. I rant. The rant has the actual thinking in it. Three or four threads tangled together, half of them more interesting than the headline I started with. By the time I'm done talking, I've been over the topic from five angles and the structure already exists. It's just not on a page yet.
That used to be where the bottleneck was. The thinking was done. Sitting down to write 800 coherent words about it took me hours I didn't have. Most of the ideas died on the rant. They never made it past the in-my-head version.
What my AI knows about how I sound
A while back I ran my entire sent-email folder through a voice-profiling pass. Years of how I actually write to clients, friends, vendors, lawyers. Sentence rhythm. Word choices I default to. Words I never use. The cadence I land on when I'm explaining something I care about versus when I'm being polite.
That profile lives in my agent's context. When I hand it a rant, it doesn't have to guess at my voice. It's already got it. The output isn't generic AI prose. It reads like an email I would actually send. Because the model behind it has read thousands of those.
What actually happens between idea and post
The loop is simple.
I rant. Either out loud transcribed, or in a brain-dump message, or as scattered notes I send to myself. The format doesn't matter. The agent only needs the thinking and the angle.
The agent turns the rant into a draft post. Same ideas. My structure. My word choices. The post lives on web-dev first. Never prod yet. The web-dev domain is locked to my home IP, so nobody else sees the draft.
I read it end to end. I make changes. Sometimes one sentence. Sometimes I rewrite half a paragraph because the angle isn't quite right. Sometimes I notice the agent put words in my mouth I'd never say and I cut them. If a paragraph reads like a generic AI-produced explainer, it gets killed.
If I like it, I push from web-dev to prod. If I don't like it, I scrap the whole post.
That last part matters. Not every rant deserves to be a blog post. Sometimes the ideas don't hold up once they're on the page. The post never makes it out.
Some people will call this cheating
Some people will look at this and say it's AI slop. I disagree. The ideas are mine. The reasoning is mine. The voice was scraped from years of me actually writing in my own voice. Every post is read end to end and edited before it ships. Posts that don't work get killed instead of polished into something I don't believe.
What changed is the friction between idea and published thought. That used to be a four-hour wall I would not climb on a busy week. Now it's a thirty-minute review. The wall moved. The thinking didn't.
The point of writing in public was never to prove I can type. It was to push ideas out. Trying to contain everything by myself, in the time I had, was costing me ideas. This system stopped costing me ideas.
Why I think this is the right shape
If you know me, yes, I'm overexcitable. Yes, I rant hard. Trying to fit everything I want to say into a coherent paragraph is genuinely hard for me. The system helps with that. The system doesn't replace me. It clears the path between what I'm trying to say and the page where it lands.
That's the whole thing. Anyone who builds AI infrastructure for a living should also be using AI to do the work that used to bottleneck them. Otherwise we're not practicing what we sell.
← Back to blog