Yes, you could do that.
Once upon a time, we did. The computer booted to a BASIC prompt, and you typed the program into the REPL, one line at a time.
Suppose you want to visit a website: there is a web browser that you could hypothetically type into your Python shell: Grail. With no ability to execute programs (because the programs can be written in the REPL instead), you'd enter the full source code of the web browser into your shell before telling it to run.
I think having to type the source code of it into your terminal each time you want to visit a website is going to get tiresome. It did for many people in the early 80s.
Instead computers developed code for loading and saving programs onto persistent storage. Punch cards, paper tape, audio cassette tape, floppy disks. You typed out your program at most once, then it was `SAVE "MYPROG",8,1` so you had it for next time. When you wanted it again, you put the disk in the drive, ran `LOAD "MYPROG",8,1`, waited for it to do its thing, then `RUN`.
For that to work, the REPL has to have the ability to load a program and execute it. This is where Python, Perl, PHP, Lua, NodeJS and co all come unstuck: they're not designed for that use case.
You can write your own, some have… but most of us are already used to the Bourne shell and its derivatives (Korn shell, Bourne Again Shell, Z-Shell). They aren't as good for scripting, but they're good enough to do a lot, and they work better as a command prompt.