I’m Maksim Danilchenko, an engineering lead based in Cyprus. This page explains how the articles here get made, because you should know that before you trust any of them.

AI writes the drafts

Articles on this site are researched and drafted by an AI agent pipeline that I built and run, using Anthropic’s Claude models. The pipeline picks topics from search and reader data, reads the primary sources, writes the draft, and runs it through a series of checks before it’s published. I set the topics, the editorial rules and the quality bar, and I’m accountable for what goes out under my name.

What each article goes through

  • Primary sources. Every article links the papers, docs, release notes and datasets it relies on, inline and in a Sources section. Numbers that come from a vendor and haven’t been independently checked are labelled “reported” or “claimed”.
  • Independent fact-check. Before anything is published, a separate agent that didn’t write the draft re-verifies the dates, numbers, versions and attributions against the sources, and flagged claims get fixed or removed.
  • Code that runs. Code in tutorials is written to be run. Since 27 September 2026, any command output shown in a new or updated article has to come from actually running it; if something can’t be run, the article says so.
  • Updates. Articles get re-checked as tools and prices change. The “Updated” date under the byline shows the last time an article materially changed.

First-person experience

When an article says “I”, it means one of three things: my opinion, something the pipeline actually ran while writing that article (with the output shown), or an experience from my own notes. Hands-on observations from other people are attributed to them.

That wasn’t always the rule. Until 27 September 2026, the pipeline was required to include a first-person “hands-on” paragraph in every article, and it wrote experience claims and test results that I can’t stand behind. On that date I changed the rule and rewrote those passages across the archive into neutral, attributed language. Where an article reported measurements nobody could reproduce, the numbers were replaced with published benchmarks, re-run on the blog’s build server where that was possible (the article then says so and shows the output), or removed. If you spot a passage that slipped through, tell me.

Corrections

If something here is wrong, email dmax.dev@gmail.com with the article link and what’s wrong. Factual errors get fixed, and the Updated date changes when they do.