MIT & Meta paper: ReHarness revises agent memory notes with later outcomes, beats baselines
rohanpaul_ai · x · 2026-11-13
A new MIT & Meta paper finds agent memory works better when old notes get corrected using what happened the next time they were used. Agents save and reuse notes from past tasks, but notes rarely change once written, so overly broad or outdated lessons keep misleading the agent.
- Method: After each task, ReHarness sends retrieved notes plus the result to a revision agent, which narrows or corrects each note using only facts from that note's original task.
- Results: Across 3 benchmarks and 3 frontier models it averaged 0.240 normalized gain vs 0.168 for the best raw-trajectory baseline; replaying full raw history used 3.4× as many tokens on CL-Bench with Claude.
Takeaway: let each outcome edit the notes that shaped your agent, not only add new notes.
Related event: MIT and Meta Propose ReHarness for Agentic Continual Learning(2 posts)→
More from coding & agent
- Rolldown ships agent-facing report mode that trimmed Framer's bundle by 1.2MB — cnakazawa · 2026-11-13
- Quinn Slack explains how Amp runs an AI-native company with 21 employees — TheTuringPost · 2026-11-13
- Pipecat 1.16 adds PerceptionReader to read prosody, tone and sarcasm in voice agents — m4rkmc · 2026-11-13
- Amp kills its Dial modes for a model selector that shows exactly who gets billed — bytebot · 2026-11-13
- Garry Tan amplifies: self-built AI code-review agent cuts costs 80%+ with better quality — garrytan · 2026-11-13
- Gemini now runs 5-hour uninterrupted coding sessions, user reports spec-compliant work — MickeySteamboat · 2026-11-13