I use AI to build and write. Not sometimes—daily. Claude, ChatGPT, local models when they make sense. It's part of my workflow the same way Git or VS Code is.
I'm not going to walk you through my setup. You can find a thousand posts about using AI for scaffolding, refactoring, research, and drafting. They all say roughly the same thing, and they're all roughly correct.
What I want to talk about is what surprised me.
I spent a decade building judgment. Learning which abstractions hold up and which collapse under pressure. Developing a feel for when code is clean and when it's just tidy. Figuring out what to build, not just how to build it.
For most of that decade, I couldn't fully use that judgment because I was also doing the typing. The boilerplate, the plumbing, the conversion of decisions into syntax. The strategic layer and the execution layer were fused, and execution ate most of the clock.
AI split them apart.
I thought this would feel like a loss — less hands-on, less craft, more management. It feels like the opposite. It feels like being promoted to creative director of my own work.
A director doesn't operate the camera, but the film is still theirs. They set the vision, make the calls, decide what works and what gets cut. The craft is in the judgment, not the execution.
I'm not writing less. I'm deciding more. What to build, what's good enough, what needs to be scrapped, what the thing is actually about. And I like that work more than I liked grinding through implementation details.
I spent a decade building the instincts to make those calls. Now I actually get to use them.
Not everyone will feel this way. Some people love the execution itself — the satisfaction of a clean implementation, the meditative quality of code flowing from your hands. That's real. But for me, the shift has been a relief, not a compromise.
Here's what I've learned not to trust.
AI optimizes for coherence, not resonance. Left alone, it produces smooth, forgettable prose — the written equivalent of stock photography. Everything is correct and nothing is alive. You have to fight it constantly to keep your voice, your edges, the specific roughness that makes writing sound like a person instead of a summary.
It doesn't know when something is a bad idea. It will confidently scaffold the wrong abstraction if that's what the pattern suggests. It has no taste — or rather, it has consensus taste, which is worse than no taste because it's harder to detect.
And it doesn't share consequences. If I ship code that breaks production, that's on me. If I publish an idea that's wrong, that's on me. The tool is irrelevant. The accountability isn't.
Before anything goes out, I ask one question: would I defend this in a room full of people who know what they're talking about? If the answer is no, it doesn't ship. Doesn't matter who wrote which sentence.
Some of my posts list an AI as co-author. That means the AI meaningfully shaped the output — proposed framings I wouldn't have reached alone, suggested structures that survived into the final piece. It doesn't mean I typed a prompt and hit publish.
The thinking is mine. I stand behind every word.
I mark the contribution because I think hiding it would be worse than any discomfort the label creates.
I tested this boundary recently with fiction.
I used AI to build the world — character backstories, plot architecture, motif tracking, chapter beats. The structural work was genuinely good. The grandmother whose Hungarian embroidery patterns become the protagonist's mathematical intuition? AI helped me find that. The engineering of a ten-chapter arc with interlocking themes? Faster and more coherent than I'd have managed alone.
Then I let it write the chapters.
The prose was competent. Every sentence was correct. No sentence was alive. The protagonist experienced anxiety through stock physical markers — her stomach dropped, her mouth went dry. The dialogue delivered exposition instead of revealing character. The descriptions were arranged for the reader's benefit rather than filtered through a specific consciousness.
It read like a very good summary of a novel rather than a novel.
The structural work survived because structure is pattern — and pattern is what AI does. The prose didn't survive because fiction lives at the sentence level, in word choice and rhythm and the tiny accumulation of decisions that become a voice. AI can approximate voice. The approximation is worse than absence because it feels dead in a way that rough-but-human writing doesn't.
It's the same lesson as photography. AI can generate a better-composed image than I'd take with my phone. It can't generate the night I froze in Poudre Canyon watching the exposure accumulate. The output is superior. The experience is empty.
In fiction, the sentence is the experience. There's nothing underneath it to point to.
I took the experiment down. The world-building stays in my notes. If I write the novel, the prose will be mine — messy, probably, but inhabited.
Some things need a witness, not a generator.
But the co-authorship question is less interesting than the one underneath it: if AI can generate infinite content, why write at all?
Not for information. AI synthesizes that on demand. The how-to post, the explainer, the consensus overview — those are commodities now.
What's left is testimony. Someone with a name, a history, and real stakes saying: I believe this, and here's why.
The value isn't in the information. It's in the fact that a specific person is willing to attach their credibility to it. It's situated — my view from somewhere, shaped by particular experiences, carrying actual risk if I'm wrong.
The test isn't "did AI touch this?" It's "could AI have written this without me?"
If yes, I haven't pushed far enough. Not louder, not more elaborate — more owned. More costly.
We're in a moment where using AI is either a secret or a flex. I don't find either posture useful.
I use AI because it lets me work at the level I've spent a decade preparing for. I don't use it as a replacement for thinking.
And the thing I keep coming back to is this: the tool changed what I spend my time on, but it didn't change what the work requires.
The work still requires judgment, taste, and the willingness to be wrong in public.
AI made the typing cheaper. It didn't make any of that easier.