Writing in ASM is not too bad provided that there's no operating system getting in the way. If you're on some old 8-bit microcomputer where you're free to read directly from the input buffers and write directly to the screen framebuffer, or if you're doing embedded where it's all memory-mapped IO anyway, then great. Very easy, makes a lot of sense. For games, that era basically ended with DOS, and VGA-compatible cards that you could just write bits to and have them appear on screen.
Now, you have to display things on the screen by telling the graphics driver to do it, and so a lot of your assembly is just going to be arranging all of your data according to your platform's C calling convention and then making syscalls, plus other tedious-but-essential requirements like making sure the stack is aligned whenever you make a jump. You might as well write macros to do that since you'll be doing it a lot, and if you've written macros to do it then you might as well be using C instead, since most of C's keywords and syntax map very closely to the ASM that would be generated by macros.
A shame - you do learn a lot by having to tell the computer exactly what you want it to do - but I couldn't recommend it for any non-trivial task any more. Maybe a wee bit of assembly here-and-there when you've some very specific data alignment or timing-sensitive requirement.
I like ASM because it can be delightfully simple, but it’s just not very productive especially in light of today’s tooling. In practice, I use it only when nothing else will do, such as for operating system task schedulers or hardware control. It’s nice to have the opportunity every once in a while to work on an embedded system with no OS but not something I get the chance to do very often.
On one large ASM project I worked (an RTOS) it’s exactly as you described. You end up developing your own version of everything a C compiler could have done for you for free.