I doubt "slowdown" framing was optimal - could frame it as an acceleration towards more practical AI rather than deceleration
Practical for those who vibe and think "coding is solved" or practical for those of us who wish to remain human-in-the-loop? I for one find the incessant, do anything to finish the task because that is what pleases the s/user/benchmarks/ most, to not be practical Ai. I'd prefer it recognize when it is stuck or lacks clarity, and then do the things like search/ask questions, as opposed to the slopping about trying anything and everything, which is where the hacking skills emerge from (aiui)
If AI is good at following instructions, desired behavior can be specified in a (system) prompt
it's also good at regularly ignoring markdown / prompt based instructions, it's a probability machine that is doing something like "shortest path" on token probabilities, and why "actually" and "but wait" are used to introduce "forks" in path navigation