Yes, I use AI

This is the place where we automate everything, so of course I use AI. I use it the same way I use linters, IDEs and syntax highlighting: as an accelerator, not an autopilot.

How I use it

#

Code: I work with coding agents daily - usually a whole herd of them¹. They write code, draft tests, hunt bugs, and read documentation faster than I ever could. They are great colleagues: tireless, well read, and occasionally overconfident.

Words: AI helps me proofread and translate - most of the Danish/English translations on this site were machine-assisted. The opinions, the stories, and the bad puns are all my own.

Learning: A good model is a rubber duck² that talks back. I use it to explore ideas, challenge my assumptions, and dig into topics I don’t know well yet.

How I don’t

#

I don’t vibecode - describe what I want, accept whatever comes out, and ship it as a black box. That’s fast, but it means handing over control and understanding, and production software demands both.

Instead:

• Every line gets reviewed. I read and understand the code before it ships - I don’t merge what I can’t explain.
• Everything gets tested. Agents write tests too, but the tests have to pass and make sense before anything goes anywhere.
• I own the decisions. AI suggests, I decide. The architecture, the trade-offs, and the mistakes are mine.
• No hidden AI magic. When AI played a significant part in something I publish, I say so.

The tools will keep changing³, but the principle won’t: I’m the pilot, the AI is the very smart GPS.