I’m still waiting for an article like this one:

https://jlouisramblings.blogspot.com/2013/04/acme-as-editor_...

It’s clear, informative, and even without reproducing it, you can still form an opinion about the tool.

All the AI pushers’s comments are either very vague or experiments whose sole quality is that it has been done with LLMs (often badly).

There’s been a few that show the Git history of a project that were (allegedly) fully written with an LLM. Not fully vibe coded, but in a meaningful amount.