There's a bunch of usecases.

For example, you can just tell Copilot to do a thing instead of spending however long it would take to google a solution. Will it gimp learning? Of course it will. But that's technology advancement in a nutshell - making things more accessible, by making them easier, thus making some baseline levels of knowledge obsolete. We no longer learn Assembly to interact with our computers, future generations won't know where Settings are, because an AI Agent will be doing it all for them.

Another example is people suffering from disabilities or even just temporary injuries. Imagine being able to fire up your PC and just tell it to play your favourite show, instead of having to use your feet or, I don't know, a pencil in your mouth, to click through the windows.

The idea is excellent. The implementation is some 5 years too early. The AI systems available, in their current form, are just not there yet for this to work well. Microsoft even pulled an ad they made recently, because people noticed that Copilot made an error there. Or even multiple errors.