interesting point, but i think I disagree a bit.
the main feature of LLMs/agents IMO is that they basically are generalized problem solvers. given that they are, in theory, built with no specific task in mind it’s fully reasonable to have the control surface be equally flexible.
obviously custom UI etc is a very cool thing, but I don’t see how people outside of IT would be happy with UI that looks different every time you use it.
replacing agent actions with scripted, fixed automations is a noble goal and clearly faster once implemented. but in a world where cost of intelligence seems to go down over time now, doing everything through agents will probably become even cheaper and thus more attractive to rely on