In a better world, AI would become a tool with the primary purpose of improving solution and code quality. Speed boosts would be a nice, yet incidental side effect. We would finally be able to ship rock-solid, nearly perfect software at a very reasonable pace, with pleasure.
In the real world, “efficient” managers – who haven’t ever come up with a better success metric than the number of (half-useless-half-bugged) features shipped in an x-week window – will never let that happen.