I use large language models. I also have strong opinions about how they should be used, and a list of things I will not hand to them. This page is where I intend to be honest with anyone who visits about where I draw my lines.
The short rule: AI helps me do work. It does not do the work for me. When a tool starts producing the thing instead of sharpening the thing I produced, I stop using it for that.
I am not interested in handing over my humanity for expediency.
What I Use It For
| Editing | Catching dumb mistakes, tightening prose I’ve already written, and arguing with me about structure. The words stay mine. |
| Research scaffolding | Surfacing leads, summarizing dense material, and pointing me at primary sources I then go read myself. |
| Code | Boilerplate, dev environment setup, and the unglamorous plumbing around projects like this site. I read everything before it ships. |
| Rubber ducking | Talking through a problem out loud when no humans are around to listen to me ramble. |
What I Don’t Use It For
| Writing | Writing is one of the ways I think. If a model writes the post, not only have I lost ownership of the ideas, but I haven’t done the kind of mental weightlifting that got me to this point. |
| Final judgment | Conclusions, recommendations, and anything I put my name behind are mine to defend. |
| Anything I can’t verify | If I can’t check and verify it myself, I don’t trust it. |
Tools
I’m building and collecting small tools — prompts, skills, and configs — that keep AI on the right side of that line: things that help me build rather than build for me. They’ll show up here as they’re ready.
(Coming soon.)