Stanford's AI agent policy debate signals a real curriculum crisis

Stanford's CS336 course, which teaches language modeling from scratch, published AI agent guidelines that got picked apart on HN. The comments ranged from 'the genie is not going back in the bottle' to 'good intention but useless.' The frustration is genuine: writing rules about AI agent use in a course that is literally about building AI systems is a philosophical contradiction that universities have not figured out how to resolve.

The deeper issue the thread surfaces is that CS education is in a weird transitional moment. The course itself, teaching LLMs from the ground up, is valuable and people are excited about it. But the scaffolding around it, the academic integrity policies, the assumptions about what a student should be able to do unaided, has not caught up. One commenter made the analogy to calculators in math class, except the calculator here can write the entire essay.

This matters for founders who hire engineers. The cohort graduating over the next three to four years learned in an environment where AI assistance was everywhere but officially discouraged in some contexts. The result is a generation of engineers with uneven fundamentals and very high AI-assisted output. You need to calibrate your hiring accordingly, because a candidate's GitHub portfolio tells you less than it used to.