AI Agent Era at Stanford: New Rules in CS336

Stanford University's computer science department has published new AI guidance for students in CS336 (Language Modelling from Scratch). Applied through configuration files such as AGENTS.md and CLAUDE.md, the rules aim to position AI tools as personal tutors rather than copy machines. Agents may not write ready-made code directly or run system commands on students' behalf; they may only guide.An AI agent is autonomous software that can decide, write code, or use tools toward a user-defined goal. In education, the step is seen as integrating AI into real life instead of banning it outright. Experts warn, however, that having AI write code directly turns students into passive observers. Struggling through debugging while coding underpins lasting learning, so tools are encouraged to use Learning Mode—offering step-by-step hints instead of handing over the solution.Cognitive load is the total effort and capacity a person spends in the mind when learning something new or solving a problem. How to enforce these rules has become a major debate. Stopping students from using other AI models in the background is nearly impossible, pushing institutions back toward traditional methods. Face-…