Maybe I'm just showing my age, but I think some of the truly impressive feats of programming are being ignored.
Steve Wozniak, Apple Disk II/RWTS
Steve designed the hardware and software together, using really cool tricks in the software to eliminate a lot of complexity (and cost) in the hardware. Normal floppy disk drives used an LED and photosensor near the hub of the disk to shine through a hole punched in the disk substrate. The output from the photosensor was connected to a processor interrupt so the processor would know when to start its timing routine to wait for the correct sector on a track to be under the read/write head (though a few used "hard-sectored disks" that had a hole to signal the start of each sector instead of just one for the start of the track).
Steve eliminated that hardware by designing the software to encode the data in a way that would let you start reading from an arbitrary spot on the disk, and not only decode the data itself, but figure out where (logically) in the track you were. The Disk II drive didn't have the LED/photsensor setup, and completely ignored the hole(s) in the floppy disk.
Gordon Letwin, HPFS
Gordon Letwin was an architect (and coder) on the OS/2 team. At least as I've heard the story, at one point he went on vacation to get away from things, and spend three weeks (or so) sailing around on his yacht (yes, early MS employees could afford things like that...) ...but to keep from getting too bored, he decided to take along his laptop.
When he returned, he had HPFS written, debugged, and working -- entirely in Intel 386 assembly language. The code he wrote was eventually sold as the "HPFS386" that was exclusive (at least at the time) to the LAN Manager Edition of OS/2. Another team then spent something like six months writing a version in C that became the "normal" HPFS included in the normal editions of OS/2 (and after IBM and MS broke up, IBM updated and re-compiled the C code to get their "HPFS386"). Although it's been modified and update since, if you were being fair about things, NTFS would probably be called "HPFS 2.0" (or maybe 3.0) -- there's no question that Microsoft's best current file system is still closely derived from what he designed.
Burroughs B220 tapes
These embodied (at least IMO) the real beginnings of object oriented programming. Where IBM tapes (for one example) had "labels" to tell about the format of data on the tape, Burroughs tapes developed a convention (I don't think anybody knows for sure who started it) of putting a small set of routines on the tape that would understand the data and how it was formatted, so you could manipulate the data correctly without knowing the details of how it was formatted. In other words, the first few "blocks" on the tape were basically a vtable in persisted form. You'd basically read the vtable into memory, then use the methods it defined to manipulate the data on the rest of the tape. All very neat and simple (if just slightly less than completely portable).