'There's no limit to how bad code can get' — and Willison's rebuttal to the rewrite-it-from-scratch response

Zach Kehs, via Simon Willison's Weblog·medium signalXBlueskyLinkedInCopy link

Zach Kehs frames the asymmetry: add floors to a building forever and it collapses, but software faces no such constraint, so there is always room for another layer of indirection or another performance regression. In the Lobsters thread Willison lays out why the burn-it-down answer almost never works — the old system stays a moving target because it runs the business, its maintainers stop investing once it is declared obsolete, and the greenfield team ships late with a subset of the behaviour because nobody ever fully understood the scope. Relevant right now because agents make the greenfield rewrite feel cheap.