I write about working with AI agents. It would be strange not to say how this site itself is made.
Who is responsible
Franz Paul holds editorial responsibility for everything published here — as a natural person, by name, with a postal address in the Impressum. Nothing goes live that I have not read and approved. If something here is wrong, it is my mistake, not a tool’s.
How the articles are made
Every non-trivial post starts as a markdown spec: a thesis, a section outline, word budgets, and a validation checklist. The spec exists before a single paragraph does. I then draft with AI assistance — Claude Code, against that spec — and run the result through a critic pass that argues with the draft rather than polishing it.
What comes out of that is a draft, not a post. I rewrite, cut, verify the claims against primary sources, and check every link. Numbers and quotes get traced back to where they came from; the ones I cannot trace get deleted. That step is the reason some posts ship weeks late and some never ship at all.
When a published piece turns out to be wrong or simply ages badly, I do not quietly edit it. I add a dated editorial note at the top and leave the original argument standing — see the MCP servers piece, where my position reversed within a few months, or the Claude Code vs. Codex comparison, whose numbers are explicitly marked as a snapshot. Corrections are part of the record, not a cleanup job.
Where AI fits
Articles on this site are produced with AI assistance and published under human editorial control. The tools draft, restructure and argue back. They do not decide what gets published, and they do not get a byline — the responsibility for every sentence is mine.
Code examples are run before they are published. Screenshots are real screens, not renderings. No image on this site is AI-generated: the illustrations are screenshots, the social cards are rendered deterministically from the post title at build time, and the animation on the homepage is a physics loop on a canvas, not a model output.