Agents can remember facts, but still forget how to do the job
No_Advertising2536 · reddit · 2026-07-21
Agent memory is often good at storing facts, but poor at storing how to do the job. This thread argues that procedural memory is the missing layer: workflows should be versioned, updated after failures, and pruned using failure history rather than success counts alone.
It cites a recent Zhejiang University + Alibaba paper, Memp (arXiv:2508.06433), which builds procedural memory from agents’ past trajectories and tests it on GPT-4o, Claude Sonnet, and Qwen. The key takeaway is that failures are the signal: the strongest mechanism was reflecting on failed runs to revise the stored procedure.
The author then lays out three production lessons:
- Session recall and workflow learning are different problems.
- Procedures need version history, not overwrites.
- A procedure that never failed is not yet trustworthy.
The uncomfortable implication is that static instruction files like CLAUDE.md or AGENTS.md rot unless humans notice failures and rewrite them.
Related event: Industry Discusses Procedural Memory for AI Agents(2 posts)→
More from coding & agent
- 250 Builders Hack With Codex at Berlin Community Hackathon Ahead of OpenAI DevDay — paw_lean · 2026-10-26
- Ben Bajarin: Frontier AI is becoming the enterprise cyber-defense stack — BenBajarin · 2026-10-26
- Building AI bots on raw APIs? Devs say models are 'dumb without the harness' — madeofAjala · 2026-10-26
- HarnessMM cuts agent tokens per task by up to 88% while raising success rates — omarsar0 · 2026-10-26
- Microsoft demos intent-classifying agents built with Copilot App and Foundry, deployable across Teams and Outlook — marlene_zw · 2026-10-26
- Looking for a wearable mic with best battery life that feeds recordings to personal agents via MCP — altryne · 2026-10-26