That is not true for AI coding at all: ie the evidence is the massive difference between Claude code in opus 4.6 vs opus 5.5

That difference is enough to unlock AI coding for so many real life teams