Markets pay thousands of analysts to guess what Jerome Powell will do — now researchers built an AI that guesses what AI-Jerome Powell will do.
The Summary
- Two researchers developed "Sim Fed," an AI simulation using agents to impersonate the 12 FOMC voting members, including digital versions of Chair Powell and Governor Cook
- The system predicts rates will hold steady when the real Federal Reserve announces Wednesday
- This is agent-based modeling moving from academic theory into real-time market prediction
The Signal
Tara Sinclair and Sophia Kazinnik built something that sounds like a gimmick but reveals where we are with AI agents: they don't need to be better than humans yet, they just need to be faster at processing the same public information. Sim Fed creates 12 distinct AI agents, each trained to think and vote like a specific Federal Reserve governor. Not just pattern-matching their past votes, but simulating their individual economic frameworks, public statements, and voting histories.
The actual innovation here is not prediction accuracy. Wall Street already has decent Fed forecasting. The innovation is speed and transparency of process. Traditional Fed watchers read speeches, parse minutes, track dot plots. This system ingests all of that simultaneously and runs the meeting simulation before the real one happens. It shows its work through agent deliberation, not just a black-box probability.
"Agents don't replace analysts here, they compress the research timeline from days to minutes."
This matters for three reasons that extend beyond whether it gets Wednesday's announcement right:
- Agent-based models can now handle complex multi-party negotiations, not just single-agent tasks
- Financial institutions will use this pattern for earnings calls, board simulations, regulatory scenario planning
- We are six months from hedge funds running agent-simulated Fed meetings as standard pre-FOMC prep
The researchers chose the Fed because it is the perfect testing ground for transparent agent reasoning. Every FOMC member publishes speeches, votes publicly, operates within known mandates. The agents have rich training data and clear constraints. This is not AGI, it is specialized simulation backed by structured data.
But here is what people are missing: if agents can simulate 12 humans deliberating over interest rates, they can simulate your executive team deliberating over capital allocation. They can simulate your board arguing about M&A. They can simulate the negotiation you are planning for next quarter. The Fed is just the most public version of a closed-door decision process.
The Implication
If Sim Fed gets Wednesday's call right, expect a wave of "AI predicted the Fed" headlines. Ignore those. The real story is that multi-agent deliberation systems work well enough now to model human institutional decision-making at scale. That means every company with meeting minutes, every board with voting records, every team with decision history is sitting on training data for their own simulation layer.
The question is not whether your competitors will build these simulations. It is whether you will build yours before they do.