ISD-Agent-Bench: A Comprehensive Benchmark for Evaluating LLM-based Instructional Design Agents

Researchers have released ISD-Agent-Bench, a massive new benchmark with 25,795 scenarios to evaluate AI agents for automating instructional design. Testing on 1,017 scenarios showed that agents combining classic educational frameworks (like ADDIE) with modern reasoning techniques outperformed all others. The study used a multi-judge protocol with diverse LLMs to avoid bias, finding theory-based agents excel at problem-centered design and aligning objectives with assessments.