Methodology
How this is made
What the process is, where AI is used, and what never ships without a person reading it first.
Process
Four steps, in this order, every time.
01
The video is the original
Everything starts as a recording: the architecture explained, then the same thing built on screen. The article is derived from it afterwards. That order is deliberate. Writing first and filming later produces a performance of work already finished; filming first keeps the mistakes in, because they happened before there was a script to hide them.
02
AI does the boring eighty percent
Drafting, restructuring, translating, and the first pass of research. It is fast and it is genuinely good at those. It is also confidently wrong about semantics often enough that nothing it produces ships unread. The judgement, the architecture and the final word are mine.
03
Nothing publishes without a human reading it
Every translation is reviewed by a person before it goes live. Publishing machine translation unread is what Google calls scaled content abuse, and it deserves the name: it fills an index with text nobody has vouched for. The database enforces this rather than trusting discipline, with a review flag per translation and no way to publish around it.
04
Claims carry sources
Statistics get a source, a publication date, and the date I read it. Where the source disagrees with the point I wanted to make, the point changes. This is also the part that turns out to matter most for being cited by AI systems, which is a convenient alignment between doing it properly and it working.
On authorship
AI is disclosed here and credited nowhere.
Every byline on this site is a person, and in practice that person is me. That is not a stance against the tools, which I use daily and talk about constantly. It is what a byline means: someone who can be asked why, who can be wrong in public, and who has to fix it. A model cannot hold any of that.
So the disclosure lives on this page, once, honestly, instead of as a badge on every article. If you want to know whether AI touched a given piece: yes, almost certainly, in the way described above. What it did not do is decide.
When something here is wrong.
It gets fixed and the fix is logged in public, with what was wrong and when it changed. Quietly editing a page is how a site stops being worth trusting.