PLAN-AND-ACTICML 2025
AI Agents Fail 9 of 10 Long Tasks — ICML Paper Splits Roles to Hit 57%
ICML 2025's PLAN-AND-ACT splits Agents into Planner + Executor, lifting WebArena-Lite success from 9.85% to 57.58%. The fix isn't smarter AI — it's st
2d ago·2 min read
