IEEE COMPSAC 2025 · QUORS Workshop · with Chung-Horng Lung
Redesigning assignments for the LLM era with theory, not vibes
Most courses either ban LLMs or improvise around them. We built a quality-oriented framework, grounded in Cognitive Load Theory and constructivism, for deliberately integrating LLMs into software-engineering assignments, and validated it by redesigning a real Operating Systems assignment.
What I did
Developed the theoretical framework, then used an LLM (Claude Sonnet 3.7) to redesign an Operating Systems assignment under the framework's constraints: phased implementation, explicit LLM-use guidelines for students, and documentation requirements. The result went through expert evaluation.
What I found
LLMs can serve both sides of the classroom: streamlining assignment redesign for instructors while giving students structured guidance that promotes metacognition, provided the integration is designed around how cognitive load actually works for novices.
Why it matters
60–80% of CS students already use AI tools, mostly without guidance. Reusable, theory-grounded assignment patterns are how curricula catch up, and how graduates arrive prepared for workplaces where AI collaboration is the norm.
The framework in three stages
Each stage constrains the next: theory decides what the LLM is allowed to change.
Cognitive Load Theory + constructivist principles define what "quality" means for an LLM-enhanced assignment.
The LLM redesigns the assignment inside those constraints: objectives, phases, LLM-use guidelines, documentation requirements.
Instructors evaluate the redesigned assignment against the quality criteria before it reaches students.
Why novice programmers get overloaded
Programming maxes out all three kinds of cognitive load at once; the framework assigns each one a different treatment. Select a load type: