Feed the system a body of text — a regulation, a news archive, a set of notes about people in your life — and it will:
- Extract the relevant actors (individuals, institutions, roles).
- Build each actor's persona from public information: position, incentives, known affiliations.
- Map the network — who reports to whom, who's aligned with whom.
- Simulate a conversation among them around a question you specify.
- Produce a report summarizing likely outcomes, and let you query any single agent or poll the group.
Cases tested
Policy: MiCA regulation (49 actors)
Fed the system the Wikipedia page on the EU's Markets in Crypto Assets regulation. It identified ~49 actors — central bank heads, regulators, business owners, journalists — and simulated how they'd publicly discuss the regulation, including modeling private vs. public positions (e.g. a regulator publicly supports their own rule but privately bets against it on Polymarket).
Personal: "Should I pivot my career?"
Using Max's notes and weekly reports, the system identified ~20 real people in his life and modeled how they'd respond to a career-pivot question. Some replies were startlingly true to character — his mother replied with something about the spring weather, exactly the way she would in real life.
What didn't work yet
Feeding the Iran–US war Wikipedia page produced ~1000 actors — too expensive and slow to simulate. The system currently scales best when the actor set is in the tens, not thousands.
Business applications
- Policy lobbying. Model a proposed EU directive, map coalition options, show a small actor what levers they actually have. (Current candidate: Convector and the chemical-ban directive.)
- Leadership 360. Take cross-evaluations from a leadership team and project "who they'd be together in 5 years" as a dynamic scenario, not a static psychometric report.
- Strategic simulation. A leader brings a decision; the system runs the likely conversation among the stakeholders they'd need to move.
Honest caveats
This is not prediction. It's structured scenario-building — useful for seeing options you missed, stress-testing a proposal, surfacing likely objections. Before any predictive claim, we need to backtest the approach on known historical events: give it the information available on a given date, and check whether the simulated outcome matches what actually happened.
Related reading
Max is working through Making Sense of Chaos by Doyne Farmer — bottom-up modeling of complex systems as the underlying intellectual frame. The same author's team built a COVID-era supply-chain model that outperformed top-down economic forecasts.