this is where ml research is headed (and honestly many other research too). nobody's replacing the thinking part, but the endless cycle of change one thing, wait, check if it worked, repeat, that eats 85% of a researcher's time and an agent can just grind through it while you