Kuaishou's AgentX-Model Uses Dual Agents for Long-Horizon Industrial Recommender Research

_reachsumit · x · 2026-09-25

A 22-person Kuaishou team published AgentX-Model, a dual-agent framework for long-horizon autonomy in industrial recommender systems research. A Research Agent forms independently reviewed proposals from papers and findings, while a Model Agent runs multi-round investigations in business-constrained sandboxes and returns code, measurements, and open questions. Research is organized around four actions—Reproduce, Follow-up, Composition, and Diagnose—so subsequent experiments build on prior results, including evidence gathering for online issues like prediction bias.

Original post →

More from coding & agent

coding & agent channel →