How I changed teaching after AI managed to do all my homework assignments

A professor who teaches an upper-level Machine Learning in Production course redesigned assessments after generative AI and coding agents routinely completed homework and quizzes. Rather than change core learning goals - engineering tradeoffs, risk mitigation, and teamwork - the course now permits AI use everywhere except written and oral exams and emphasizes responsible tool use. The instructor argues AI undermines evidence-backed practices like frequent low-stakes assessments and resubmission safety nets, so some practices were altered (e.g., a 10% penalty on resubmissions) to discourage outsourcing learning while preserving opportunities for students who want to engage deeply.

Concrete changes shift assessment weight and verification into supervised, in-person interactions and artifacts. Written reflections and online reading quizzes were largely abandoned in favor of 15-minute TA check-ins (pass/fail, with retries) that probe understanding, plus team debriefs and short video demos that require functioning features. Labs and assignments require demonstrable evidence and live knowledge checks. Autograding via institution-approved LLMs flags pass/needs-review results, cutting TA grading time by roughly half to 80% and freeing TAs for face-to-face work. Exams and classroom activities now carry more grade weight, and some accommodations (in-person only checks, bonus points for deep explanations) help ensure students actually understand their submissions.