Meta paper: Agents must learn when to use tools, not just have them, achieving 96.9% on ALFWorld
daniel_mac8 · x · 2026-08-24
A new Meta paper suggests that simply giving an agent more memory and tools is insufficient; the agent must learn when they are worth using. The proposed EvoHarness-RL method achieved a 96.9% success rate on ALFWorld using Qwen2.5-8B, while reducing harness usage to about 1 call per episode, addressing state management and forgetfulness in long-running tasks.
Related event: Meta's EvoHarness-RL Trains Agents to Decide When to Use Memory(2 posts)→
More from coding & agent
- Gartner: 75% of Organizations Saw Unauthorized AI Coding Assistant Use — moniquejmorrow · 2026-10-16
- Photon 2.0 launches unified messaging API for AI agents across iMessage and more — Scobleizer · 2026-10-16
- Open Nshipyard cleans Canada's messy civic data, exposing $15.1B phantom spend, with MCP tools for agents — RichardsonDx · 2026-10-16
- Claude Opus 5.5 effort levels tested: low reasoning iterates 9x faster than max — tom_doerr · 2026-10-16
- Figma launches Guidelines and Code Sources so AI agents can follow your design system — JaynitMakwana · 2026-10-16
- Ex-DeepMind Eng Lead's AI Calling Agent Fo Cut Hang-Up Rate from 80% to 20-30% — julianweisser · 2026-10-16