> it suggests that these tools might actually be better at delivering the perception to the user of increased productivity while real productivity gains are lower, or maybe zero, or maybe negative in some cases.

Or that the gains are in niche applications (like the GP's) that don't translate at scale.