I think also the key is - don't call it AI, because it's not. It's LLM assist query parsing and code generation. Semantically, if you call it AI, the public expects a cognitive equivalent to a human which this is not, and from what @tptacek describes, is not meant to be - the reasoning and other code bits to create agents and such seem to be developed specifically for code generation and programming assist and other tasks thereof. Viewed in that lens, the article is correct - it is by all means a major step forward.