UMich's 15-week agent engineering course makes students build a coding agent and explain why it fails
kalyan_kpl · x · 2026-10-19
University of Michigan's EECS 498-016 "Applied Agentic Software Engineering" is a 15-week course built around one coding agent, with lecture notes, slides and demos being open-sourced.
- Structure: Weeks 1-3 use Aider with a local small model (ollama/qwen3.5:9b) to learn the "Big Three" failure diagnosis—context, model, prompt; students then drop below the abstraction and build their own agent against a raw LLM API with tool calls and the loop; the back half adds memory, evals, skills, permission gating and safety, ending in a deployment.
- Experiments: Students design at least 8 test tasks, run each 3 times, and measure success rate, cost and latency, then analyze failures—"take the human out of its loop and measure what broke."
- Grading: Create 49.5%, Analyze 22.5%, Apply 18%; no exams—projects, labs and demos only.
Recommended for developers already using coding agents who want to truly understand agent engineering.
More from coding & agent
- Switching to Codex GPT-6.1: dev reports confidently wrong answers on large codebases — ankitjangidx · 2026-10-19
- How Mission Pandora keeps viewers watching a 35-minute AI film: log-entry storytelling — lmoroney · 2026-10-19
- BELIEFWEAVER: belief updates and world modeling for long-horizon agent failure recovery — IntelligenceLab · 2026-10-19
- DeepMind's XYEval: one bad user hint drops agent success from 67.4% to 36% — i_dg23 · 2026-10-19
- Pure three.js + Claude/Codex Yields Fully Procedural Game Characters — TinfoilTricorn · 2026-10-19
- Open-source Kapitsa: a Metal Engine for Apple Silicon Pushing Qwen 27B to 63 tok/s on M4 Pro — DotaMate · 2026-10-19