Five years ago, you handed AI a plan.

Two years ago, an experiment plan.

Now you hand it a problem.

Each shift didn't make the previous step faster.

It changed who designs the experiments.

That distinction is the whole ballgame.