Meta's A-MLE paper: 39 authors, agents tackle the ML iteration bottleneck in ads ranking
rohanpaul_ai · x · 2026-09-22
The A-MLE (Agentic ML Exploration) paper (arXiv:2609.08248), authored by Erwin Gao and 38 others at Meta, lays out the system behind the agent framework deployed on ads-ranking models.
Core thesis: modern industrial ads ranking stacks are bottlenecked not by model capacity or training compute, but by the throughput of human ML iteration—each cycle of research, implementation, training, debugging, evaluation, and launch takes days to weeks of senior engineer time per model. With many heterogeneous models in a stack, effective techniques diffuse slowly, leaving substantial recoverable signal unexplored.
A-MLE decomposes ML iteration into five orchestrated stages: hypothesis generation, exploration strategy, experiment execution, result analysis, and a shared knowledge substrate, letting autonomous LLM agents systematically explore ML techniques across the whole model portfolio.
Related event: Meta Open-Sources A-MLE: LLM Agents Automate Ad Ranking ML Experiments(2 posts)→
More from coding & agent
- TypeSafe AI launches JEV, a decision model hitting 70-500ms at $0.042/M input tokens — sven_ai · 2026-09-22
- Google's RRSI regularizes recursive self-improvement of agent harnesses, +4.7 OOD points — google · 2026-09-22
- 53 harness bugs: an engineering team's field notes on building eval harnesses — DrDatta_AIIMS · 2026-09-22
- atuin 18.23 ships infinite searchable scrollback — shell history now keeps command output — aronchick · 2026-09-22
- Kev: open-source 0.8B/4B/9B judge models on Qwen3.5, the 9B fits a 32GB Mac — khiladi1729 · 2026-09-22
- Open-source coding-agent skills for Google ADK on Google Cloud, distilled from real engineering lessons — Difficult_Design6676 · 2026-09-22