Evaluate AI Coding Agents by Task, Not by One Score

A task-stratified evaluation helps teams see where coding agents perform well and measure review effort, rework, and acceptance over time.