I spent multiple years of my life working on a (at the time I started) decade+ project to retire such a system (i.e. a duct-taped hodgepodge of poorly documente

I spent multiple years of my life working on a (at the time I started) decade+ project to retire such a system (i.e. a duct-taped hodgepodge of poorly documented COBOL and mainframe assembler that was understood and maintained by staff who was entering retirement age at an astounding rate), at a "midsize" financial company (i.e. between 3k and 4k employees, all told). While the cost of maintaining the old system was certainly high, it worked astonishingly well (provided you didn't need to add new features, such as a "graphical interface"), and the cost and difficulty in replacing it was truly enormous. I think you're drastically underselling the difficulty of re-implementing a company's entire suite of LOB software, complete with migrations and training, and likely with very little in the way of flexibility to change the core business rules/logic. Furthermore, > they're held together with a fragile weave of scripts, hacks, unsupported hardware and outdated knowledge Swapping out COBOL for typescript and mainframes for linux boxes and a pile of YAML is not going to change any of that. People can and will build garbage with any set of tools you give them. Not ever software problem is like use-after-free where you can fix it forever with a sufficiently clever piece of tech.