Of course they do, but typing stuff is just physically slower for some operations. Typing git log, then scrolling through the pages of text in order to find the interesting change will always be slower than a right click -> log and then "pointing at things".

I think your conversation metaphor is good, but in a GUI the answer can be richer and interactive. A CLI can only manipulate text, a GUI can manipulate text where necessary and use a better medium (images, graphs, tables) where necessary.

There are always advantages and disadvantages of both approaches. I always shudder when I recollect configuring IIS using GUI ;).

CLI does not need to manipulate text only necessarily, see the command line in AutoCAD for example.

But yes, GUI in many cases is a better way, I wouldn't want to have cli-only Photoshop-like app (or even cli-only CAD modeller). The point is to recognize effectivity of both approaches for a given problem, at the given abstraction level.

Would you rather have a conversation with your car "left 20 degrees, now right, slow down a little, whats the speed atm?" or just turn the wheel and press some pedals?

That's not a good example - the conversation with the car would be "take me to {someplace}", which couldn't necessarily be a specific name of the place, it could be "where I met with {someone} {sometime} ago". Think Star Trek conversation with the computer ;).

There would be no point in micromanaging the car using a limited vocabulary.

But I can't go to command line and tell it "delete all of the files from the project X I do not want anymore" or "find me the change in this file in the history that caused bug Y".

Sometimes it is better to mass delete files with a find, grep and rm, sometimes an auto-filtered search and cmd+a cmd+delete is better.