I believe the main size optimization here comes from having only teletype and batch processing interfaces. When you do that, the surface area of the UI, and hence the amount of supporting code, drops tremendously. These early tools also skimped on error messages and checks, so the user experience was generally one of confusion and catastrophic error, despite being small and simple. You really needed the documentation to have any hope of understanding an old system.

Now we have operating systems that go out of their way to automate away everything and present it in real-time with multitasking and custom graphical elements everywhere, while also supporting many more protocols and hardware interfaces - wireless networking, GPU APIs, etc. The biggest growing pains seem to be past - things went from simple and stable to complex and unstable in the 1990's and then to complex and stable but insecure now. There's a lot of room to mature all of these features, but there aren't as many novel ones.

One reason is that .COM programs were copied to memory as they were, in a single segment, and therefore could access a max of 64 kilobytes of RAM, and they itself could be max 64 kb. They are a tiny memory dump of the program.

.EXE programs have loaders that handled the segmented memory and could access much more, up to the 1MB of memory space. That loader of course made .EXE programs larger than .COM.

> could access a max of 64 kilobytes of RAM

This is incorrect. A COM program had to fit its code and static data into 64K, but it could access as much RAM as existed on the machine.

A related misconception is that the segment registers in COM programs had to have all the same value. This is incorrect as well, as the program once loaded could reset them to any value.

Indeed! First thing the executables generated by my early DOS compilers did was to set up the es: and ds: segments in order to get more space.

Eventually I overcame laziness though, and started to study the EXE file format, which, AFAIR, does not have a "loader", (as Shorel suggested) but just a header describing the layout of the file. I think EXE files tend to be larger, just because they can be larger.

MZ (DOS .exe) requires a loader - smarter than just "copy this whole thing into memory and jump to start", anyway - because it has a relocation table. But, of course, the loader is not part of the executable itself, so it doesn't contribute to its size. I suspect that the reference was to the relocation tables themselves.

> and they itself could be max 64 kb

Minus 256 bytes at the beginning for the Program Segment Prefix (hence why every .com assembly listing usually starts with ORG 100h).