Students are shy to answer questions. In stead of waiting, I pass around a stuff animal, so whoever's holding it gets to answer. This also helps the student be ready when it's their turn. If the class is split in-person/zoom, I attach a wireless earbud to the stuff animal, making a poor man's catchbox.
Instead of making an AI use policy for homeworks, I don't grade homeworks, but give them quizzes that will be easy if they have done their homework, so they are self-motivated to actually do the homework.
I make slides, but don't show them in class; the slides guide me when I'm writing on the blackboard, and also function as lecture notes for students to review. The blackboard slows me down and the students follow better.
I make slides in markdown and convert them to PDF with pandoc. This stops me from wasting time fiddling with layout, and makes it easier to get help from AI. With pandoc I can also convert the same md source into a web page with interactive code snippets that the students can play with.
I use a somewhat unusual grading scheme for multiple-choice questions where the correct answer selects multiple options (at least it's not natively supported by GradeScope as I write this! But it's still easy to implement the scheme in GradeScope by editing the rubrics): I take off points based on the Hamming distance of the student answer and the solution. In other words, each bubble worths 1 point, and the student gets the point if they correctly fills it/leaves it blank. This way, everyone can get fine-grained partial credits in a fair way.
We literally read the paper/book in my reading seminars.