This article seems like LinkedIn style controversy-bait, or is written from a place of extreme pedantry.

Evals can come pretty close to getting “deterministic” output from LLMs, and I’d argue that this is reasonably considered engineering.