The problem is that for any of these kinds of applications it is possible to do either a great GUI, which enables maximum user productivity, or a bad GUI, which slows down the user.
Sadly, if I try to remember the best user interfaces for such programs, the award still goes to some ancient MS-DOS programs that I had been using 30 years ago, i.e. the Brief editor for programmers, the Xtree Gold file manager and the Lotus 1-2-3 spreadsheet.
Those programs did not have real GUIs, because they had to use the text user interface of MS-DOS, but they already had the most important features of any GUI, e.g. tiled windows, pop-up windows, menus, histories for any kind of user input, many kinds of visual selection, scripting, keyboard macros, and so on.
After you learned the keyboard shortcuts for menu commands and maybe you also developed your own custom keyboard macros, it was possible to perform any task on those ancient programs much faster than on any modern equivalent.
I am not sure which is the reason, but maybe those old programs were more efficient because the mouse was optional. You could use a mouse if you had one, but the programs were perfectly usable without a mouse.
I assume that because of that, the developers paid a lot of attention of how to ensure that every task could be done by pressing a minimum of keys.
Even if most modern GUIs are supposed to also support a keyboard-only use, in practice I always discover that some operations are either impossible or extremely awkward when not using the mouse. (when using just the keyboard, a graphic selection cannot be expected to work as well as with a mouse, but any kind of text selection should be as fast or faster)
A point-and-click user interface is supposed to be better for a casual user, who cannot be familiar with the complex keyboard-shortcut sequences that might be needed for performing a task at maximum speed. This was certainly true for almost all GUIs in the first one or two decades of their use, but many modern GUIs have a lot of very well hidden and hard to discover options, so being a GUI becomes no improvement over a CLI where you need to read the man page.
In conclusion, there is no doubt that GUIs should be better for many tasks, but in practice one encounters too many bad GUIs, for which it may be simpler to use ancient CLI alternatives.
I think you've never met a LaTeX fan.
Regarding spreadsheets and file managers: for simple tasks I agree they can be better than the command line. But for even a moderately complex spreadsheet I usually prefer a python script, and I've seen office workers spend literal hours doing something that could take me under a minute using the command line.
Word processors: are a real improvement over using ed at the CLI, or really any scripting tool, to process large amounts of human-readable text. Although in many cases, simple markup on simple text in a simple text file can be excellent.
Spreadsheets: for all that people complain about spreadsheets, they do a few things fairly well (cellular data model, dependency-graph-driven computations)
PowerPoint: I should have mentioned that "sheets/slides/presentation" apps are one of the best examples of this; features like "Align vertically" are a great example of where UIs make certain operations very intuitive and easy
Browser: I dunno, I prefer e-links :)
Music editing: Actually, the whole world of track-based video and music editing. This is a such a huge improvement over any other system (musicians and video professionals may disagree), specifically the ability to quickly view, cut and paste, and combine multiple parts of longer videos. A lot of this depends on a concept that only really became feasible once disk storage really became huge: https://en.wikipedia.org/wiki/Non-linear_editing
Composability and orthogonality go hand in hand- much of my argument rests on the idea that composing orthogonal features is trivial for UIs, whereas composing interacting features leads to combinatorial complexity and can really only be expressed in code (and possible visual programming- being used more and more commonly to give artists GUI control over complex code-like operations).
The automation is the last interesting bit- will visual programming ever become ubiquitious, so that most people doing automation don't actually drop down to some CLI/text interface/programming language?