RD-AGENT Lecture 1: How AI Factor Factories Work
Lecture 1: How the AI Factor Factory Operates Learning Objectives: Understand the five steps of the RD-Agent class factor mining cycle, the division of labor between humans and AI, and be able to read the execution log of one cycle. 1. Introduction: A Real Machine That Has Been Running All the materials for this course come from a real battle: in September 2026, on a futures panel of 23 varieties (from 2022-12 to 2026-09, approximately 918 trading days), using RD-Agent to automatically mine factors in a loop. The first round (13 varieties) produced 136 experiments; from the second round to the first window closure, 40 cycles were run, producing over 200 work areas. Here is the conclusion, which will be demonstrated in every lecture of this course: The AI factor factory replaces human "hands", not human "judgment". It can write code for dozens of factors and complete backtesting in one night, which is true; however, whether the proposed hypothesis is valid, whether the results can be trusted, and the judgment of each link are still done by humans - and whether the results can be trusted is precisely the most easily mistaken link.
An AI factor factory was used in a real campaign in September 2026, involving 23 futures varieties over approximately 918 trading days. The AI system, called RD-Agent, automatically mined factors in a five-step cycle: idea generation, coding, backtesting, feedback, and recording. The AI's role was to generate and implement factor assumptions, while humans were responsible for data selection, testing parameters, and verifying results.
The AI produced over 290 factors, with the top one reporting a 26% excess return, but these results require further verification.
Written by urgent.news from Dev.to's report — not a translation of it. Machine-written — may contain errors; check the original before relying on it.
