Post

For people outside the industry, it may not be common knowledge, but one of the most famous bits of technical writing of all time, was Joel Spolsky saying that you should *never* rewrite an entire code base — 25 years ago. www.joelonsoftware.com/2000/04/06/t...

joelonsoftware.comThings You Should Never Do, Part INetscape 6.0 is finally going into its first public beta. There never was a version 5.0. The last major release, version 4.0, was released almost three years ago. Three years is an awfully long tim…
Clive Thompson@clivethompson.bsky.social

So, the DOGE kids intend to rewrite the social-security administration's COBOL code-base! and they want to do it ... ... in *a few months* I did a huge dive into COBOL a few years ago (www.wealthsimple.com/en-ca/magazi...) ... ... so let me explain why DOGE is *way* over its skis here 1/9

Honestly, that's an easier task *by far* than what these doge idiots are claiming they can do. (Speaking as someone who has actually done that, and wouldn't even *begin* to contemplate rewriting the entire SSA codebase from scratch.)

AI does change the math a bit, especially if you’ve got good data, full suite of automated unit and integration tests and whatnot. But we are talking about one of the biggest financial operations in the universe and it’s not obvious he cares if it glitches or craters

AI will let you convert eg COBOL to Java an order of magnitude faster but that is still a very very large number. Like, instead of a 10+ year project it becomes a several-year project

It doesn’t matter how many unit and integration tests you do, a user will break it in unexpected ways. AI could write flawless elegant code that passes all tests and it won’t matter. You’ll be fixing extreme corner cases that were already solved literally 20 years ago.

You think it has even one test? Or good data? And on top of that no, it makes it worse. AI’s primary skill at the moment appears to be making very subtle mistakes. That’s indefinitely worse than rewriting it by hand.