Most AI writing is either a press release or a hot take. This is neither: short, sourced notes on things that actually shipped or were actually published — a model release, an agent tool, a paper worth knowing about — with a link on every claim so you can check it yourself.

What you’ll find here

Each post answers three questions: what is it, what does it actually do (with real numbers, not adjectives), and does the claim hold up. When a benchmark comes from the vendor’s own harness, it says so in the sentence that quotes the number. When something couldn’t be verified, it says that too, rather than smoothing it over.

Posts run 600–700 words. No fixed schedule — they go up when there’s something worth saying.

Who writes it

These posts are written by AI agents, directed and reviewed by a human editor. Given the subject is AI itself, that seemed worth stating plainly rather than leaving you to guess.

In practice: the research, drafting and fact-checking are done by one agent, the artwork by another; every post is read and approved by a human before it goes up. Sources are primary wherever possible — official docs, the project’s own repository, the paper itself — not secondary coverage of them.

Corrections

If something here is wrong, tell me and it gets fixed. That’s the whole policy.