A defence of shitty scientific programming. I'd like to outline some of the issues that scientists learning to code on the job might face, and also acknowledge how learning some software engineering skills will help your science progress faster and more reproducibly, with a few (not original) ideas about what those helpful things might be.