Agents enter the quant research process. Execution remains deterministic.
A reference architecture for integrating agentic AI into quantitative investment research, balancing autonomous research with deterministic execution.
Paper
Agentic AI for Quantitative Investment Research: A Practitioner's Guide
Authors
Anna-Helena Mihov and Philip Treleaven
Date
13 November 2026
Most approaches to agentic quantitative research face an architectural trade-off: a single general-purpose agent handling the entire workflow, or multiple specialised agents coordinating individual tasks. This paper surveys more than 60 published agentic research systems and proposes an alternative.
The proposed architecture separates research planning and evaluation from a deterministic execution core. Agents handle tasks such as hypothesis generation, experiment design and interpretation of results, while calculations and backtesting remain controlled and reproducible.
The authors also address research integrity, including data snooping, experiment tracking and the need to define test designs before execution.
The paper provides a practical framework for deploying agents across the quantitative research process without sacrificing the reproducibility and control required for investment research.
