Who we are & why

AI Chronicle began in 2024 as a private newsletter to about fifty addresses: every week Alexey Voronin distilled the AI news stream for colleagues — without press-release enthusiasm and without panic headlines. By year's end the list had grown to several thousand subscribers, and the real gap became obvious: not news, but verification. Numbers from vendor decks travelled through the media without a single attempt to reproduce them.

That gap produced the rankings with control runs — first a table inside the email, then a standalone page, and the site grew around it. In 2025 Maria Sokolova joined to own the testing methodology, and Dmitry Belov brought a case library from consulting practice. The English version launched the same year for a simple reason: half the primary sources and half the readers live outside the Russian-speaking space.

We measure ourselves by one metric that never appears in analytics dashboards: whether a piece still holds up when re-read six months later. Chasing that standard is slower — and it is the entire point.

The name reflects the method: we keep a chronicle rather than chase lightning. Every industry event lands on the site with context — what preceded it, what it changes and for whom. The format is slower than news feeds, but readers name it as the reason they stay: it saves not five minutes on a headline but five hours of assembling the picture themselves.

Today it remains a compact four-person newsroom by deliberate choice: grow slowly, take no money from the subjects of our coverage, and keep every number on the site reproducible. Plans for 2026: an open dataset of our control runs and a public archive of every correction.

Three promises to the reader. No pieces "based on a press release" — only what we verified hands-on or confirmed with a primary source. No hidden integrations: if a text is paid for, it says so above the headline. And no fear of boring conclusions — when the honest answer is "the difference between models is invisible on your task", that is exactly what we publish, even when a louder headline would collect more clicks.

A piece travels this path: topic from the editorial backlog → brief with theses and sources → author's draft → fact-check by a second person (every figure verified against primary sources) → publication with a date → "Updated" notes for any substantive change. An average article takes four days to two weeks; the rankings run on a separate pipeline with API control runs.

On AI in our own work, plainly: we use models for interview transcripts, structural drafts and proofreading — it speeds up the routine. The final text, every conclusion and every fact are the work and responsibility of humans. There are no "fully AI-written" pieces on this site and there will be none — that would contradict the point of the project.

Money: the project lives on reader subscriptions and labeled ads from companies outside the AI industry. That structure lets us write about any model's weaknesses without looking over our shoulder — which the rankings regularly do.