Michael "Big Balls" Bolton rewriting SSA code
https://www.wired.com/story/doge-rebuild-social-security-administration-cobol-benefits/
For reference, SSA has roughly twice as many lines of code written in COBOL as the Linux kernel has lines of code written in C.
I'm part of a project that does something extremely close to this in Germany. After two or three years we've barely scratched the surface. The complexity of such a system with the wide range of rules, regulations and best practices is mind bogging. Just reverse engineering the legacy system alone takes years.
I don't envy you. I learned cool in university and vowed to never touch it again. Without going into too much detail, what are some of the common problems?
It's the same problem as with any other software development: Politics (literally). Some decisions are made by people who are not qualified to make them. Because of the scale of the project, these decisions affect hundreds of devs across dozens of teams with millions of euros swinging one way or the other. Apart from that, when divide and conquer is done properly, the work of each individual team isn't too different compared to software development in commercial companies. Everything is a bit more relaxed, though. That can be a boon but can also be infuriating if you're waiting for licenses, hardware, or some team to act.
Almost a century of history is coded into this system, it's ridiculous that they think they can just rewrite it.
Oh, it can be rewritten and it should be. But it's a very time consuming and expensive process. It may take a long time to amortize indirectly though a system that is more stable, easier to maintain, adapt to new laws, and requires less effort to use and operate. It's not a quick win and not a win in the time frame of one administration - which is why it's usually not considered a high priority.
Even if you can do this with AI and it produces a fully working result, the biggest problems in legacy systems are not the programming languages they are written in.
In fact converting one programming language into another doesn't require AI. That's because programming languages are uniform and exact. You can easily find tools that do this already.
The biggest problem is decades of technical debt and forgotten rules. With old mainframes the people who wrote the code are often no longer with us.
Will it be like Office Space where the rounded off cents goes to an offshore slush fund owned by a shell corporation?
Will there be conjugal visits?
Come on, this isn't Riyadh. They're not gonna saw your hands off, all right? The worst they can do is put you for a couple of months into a while collar, minimum security resort!