I think it's really sad that programming assignments have basically become untenable in the era of LLM/coding agents. I'm a theoretician in my research but I learned so much more by actually writing code and debugging than I did from topics that I only saw in lecture. Exams make a lot more sense to me for an actual theory course rather than compilers class.
If I deploy projects that were designed for a pre agent world, then students can basically use the LLM to get a perfect score unless I play a losing cat and mouse game of LLM police. And the students who don't use LLMs are at a huge disadvantage score-wise.