Teaching in the Age of AI: Moving from Written Homework to In-Person Verification

In response to AI's ability to solve traditional homework, this CMU course was redesigned to prioritize in-person oral check-ins and video demonstrations over written assignments. The instructor increased the scope of coding tasks to massive, real-world codebases that challenge current AI agents while shifting more grading weight to supervised exams. By using AI to automate routine grading, the teaching staff can focus more on direct student interaction and verifying conceptual mastery.

Key Points

• AI agents have made traditional written reflections and small-scale coding assignments obsolete as measures of student understanding.
• Assessment has shifted toward high-touch, in-person interactions, such as oral exams and TA check-ins, to verify genuine learning.
• Assignment complexity must be significantly increased to production-scale environments to challenge the current capabilities of AI coding tools.
• AI can be leveraged by educators to automate routine grading and provide instant feedback, allowing human instructors to focus on student engagement.
• Continuous, semester-by-semester adaptation of curriculum is now necessary to keep pace with the rapid evolution of AI tools.

Sentiment

The community is mostly sympathetic to the article's premise and broadly agrees that AI has made conventional homework unreliable as an assessment signal. The discussion is not anti-AI so much as worried about incentives, student motivation, and the loss of productive struggle. Dissenters argue for more AI-native teaching models or criticize universities' credentialing role, but the dominant view favors practical redesign around live demonstrations of mastery.

In Agreement

• Traditional homework can no longer be trusted as evidence of individual understanding because students can outsource the work to AI while learning very little.
• Homework remains valuable as practice, but the actual grade signal should come from in-class quizzes, exams, oral checks, or demonstrations where students must explain their reasoning.
• Evaluation is part of teaching because it provides feedback, motivation, and guardrails against students taking shortcuts that undermine their own learning.
• Completion-based homework combined with immediate in-person assessment can preserve practice while reducing incentives to submit AI-generated work.
• Students need explicit exposure to AI's confident mistakes so they learn not to treat generated answers as authoritative.

Opposed

• Some commenters argue that AI improves teaching and only breaks outdated evaluation formats, so the article overstates the teaching crisis.
• Some believe universities should stop requiring graded homework entirely and shift toward optional practice plus exams, instead of trying to preserve old assignment structures.
• A more radical view holds that AI exposes higher education as mostly credentialing, money, and control, and that AI tutors or AI-led assessment could replace much of the professor role.
• Some skepticism focuses on whether frequent oral or in-person verification can scale fairly without excessive instructor workload or institutional rigidity.
• A few commenters suggest students should be trusted or motivated intrinsically, pushing back against claims that grades and assessments are necessary for learning.