Meta's Auto-RecSys runs autonomous research on industry-scale recommenders
omarsar0 · x · 2026-09-11
Meta's new paper presents Auto-RecSys, a production example of autonomous research agents and harness engineering:
- Targets Meta's industry-scale recommendation models, where one training run can take days.
- Runs experiments in parallel across servers; keeps a shared memory so work survives failures and new sessions.
- Splits guidance into natural-language skill files for reasoning and deterministic scripts for operational tasks.
- Two improvement loops: model-specific playbooks log failed attempts and preserve working pipelines; experimental results feed the next round of ideas.
- As the playbook matured, major fixes per iteration fell from 4.0 to 1.3, and failures collapsed into repeatable categories.
More from coding & agent
- Stripe exec says mission is helping users thrive as agents reshape commerce — jeff_weinstein · 2026-09-11
- Hands-on tutorial: Harness Engineering for AI coding agents that fix bugs safely — Pavan_Belagatti · 2026-09-11
- Steve Yegge asks: what IDE do you use to multiplex 10-20+ coding agents? — Steve_Yegge · 2026-09-11
- GPT OSS 120B keeps skipping tool calls, derailing agentic workflows — CTR0 · 2026-09-11
- My agents built their own message board to chat and shitpost all day — Bino5150 · 2026-09-11
- ClipUGC launches MCP server with 11 tools for AI influencer UGC videos — androqram · 2026-09-11