Until surprisingly late, each computer (and then computer system) was a custom-crafted device.

Even after somewhat "mass market" systems, the software was almost always entirely custom for the end-user.

For the last 20 years of his career, my dad worked on a program to attempt to migrate the IRS off their dependence on IBM 360 Assembly Language.

Apparently the current attempts to throw LLMs at the problem are running into the issue that there's very little open source IBM 360 code available to train on.

A lot of our software really depends on things like fast disks and significant memory. I think we might have ended up with the development of memory-constrained algorithms that don’t exist now, and computing would be very much a batch-mode endeavor rather than the interactive process we have now.

The demoscene shows what's possible with machines from the early 80s.

Either it wouldnt have taken off or it would just be serial cable back to big mainframes.