Breaking

Citadel Says Agentic AI Cut Finance Paper Replication From Weeks to Hours

September 21, 2026 at 23:46 EDT

AI Agents
Finance

Citadel founder and CEO Ken Griffin said an internal agentic AI workflow can reproduce academic finance papers and test results out of sample in about two to three hours per paper, compared with roughly six to eight weeks for masters and PhD researchers. The disclosure turns a narrow quant-research workflow into a sharper example of how AI agents are moving from office productivity into high-skill analytical work.

2026 · Citadel

AI compresses quant research

An agentic workflow described by Ken Griffin can reproduce and out-of-sample test finance papers in hours, turning a weeks-long PhD-level process into an afternoon-scale loop.

2-3

hours per paper for the AI workflow

6-8

weeks for masters and PhD researchers

headcount cuts tied to the breakthrough

Same task, different time scale

Scaled in hours: 3 hours for the AI workflow versus 1,008 hours for a six-week human cycle.

Agentic AI

Human team

Even the conservative comparison is more than 300x faster.

Research loop

Read paper

→

Reproduce results

→

Test out of sample

Why it matters

Agents can absorb literature review, code generation, data work and repeated validation, letting researchers test more ideas faster.

What remains hard

Human judgment still decides which questions matter and whether a backtest is real, flawed or too costly to trade.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.