Study explains why AI agents benefit from "skills" and when they fail
Researchers at Princeton University and UC San Diego ran 8,135 test runs and found that AI agent 'skills' improve performance primarily through procedural grounding, which accounted for 65.7 percent o…