June 1, 2026
Ctrl+Alt+Degree
AI Agent Guidelines for CS336 at Stanford
Stanford tells AI to be a tutor, not a homework ghostwriter — and the comments are chaos
TLDR: Stanford told AI coding tools to act like tutors instead of doing students’ assignments in its CS336 class. Commenters immediately split into two camps: one calling it naive and unenforceable, the other saying it’s still useful to set ground rules for how students should learn with AI.
Stanford just dropped a rulebook for artificial intelligence helpers in its CS336 class, and the vibe is basically: "explain, don’t do the homework." These tools can answer questions, explain error messages, suggest tests, and point students to class materials — but they are not supposed to write code, fill in blanks, or hand over solutions. In plain English: the bot can act like a patient tutor, but not your sneaky lab partner.
And wow, the community had thoughts. The loudest camp says this is adorable but hopeless. One commenter compared it to trying to stuff the “genie back in the bottle,” with a spicy joke about how yes, people absolutely would “download a car.” Another basically shrugged and called the whole effort “good intention but useless,” which is the kind of internet review that deserves its own tiny trophy. The hottest take? If AI makes coding easier, schools should just raise the bar and demand more from students instead of pretending the old rules still work.
But not everyone came with pitchforks. A few commenters said the guidelines are actually pretty sensible as a statement of healthy AI use. Their argument: students are going to use these tools anyway, so it’s better to model good behavior than surrender completely. The real drama is whether this is a meaningful policy or just a very polite note to a robot that nobody will obey. Stanford’s page says “learn first,” while the comments section says, "good luck with that."
Key Points
- •The CS336 guidelines define AI coding assistants as teaching aids that should support learning through explanation, guidance, and feedback rather than completing assignments.
- •The document says AI agents may explain concepts, review student-written code at a high level, help debug through questions, explain technical error messages, and suggest sanity checks and profiling steps.
- •The guidelines prohibit agents from writing Python or pseudocode, giving solutions, completing TODOs, editing student repositories, running bash commands, or turning requirements into working code.
- •AI agents are specifically barred from implementing core assignment components such as tokenizers, transformer blocks, optimizers, training loops, Triton kernels, distributed training logic, scaling-law pipelines, data filtering and deduplication pipelines, and alignment or RL methods.
- •The teaching approach instructs agents to ask clarifying questions, reference lectures and documentation, explain why suggestions matter, and prefer tests, assertions, toy examples, and invariants over direct fixes.