I’ll share more detail soon, but here’s what surprised me.
In the first version, the agents were simple: voting records, public statements, committee assignments, donor data, and constituent polling. Then I let them vote.
Bills that have stalled for decades sailed through. Getting money out of elections passed with 95% approval. Banning insider trading for members of Congress hit 98%. Ending continuing resolutions cleared 95%.
Turns out, if you model lawmakers as mostly rational actors responding to inputs, they behave… rationally.
To make it realistic, I had to add self-preservation, incentives, and a strong sense of “the other side.”
Now those same bills die in committee. As expected.
I built an end-to-end simulation of Congress.
I’ll share more detail soon, but here’s what surprised me.
In the first version, the agents were simple: voting records, public statements, committee assignments, donor data, and constituent polling. Then I let them vote.
Bills that have stalled for decades sailed through. Getting money out of elections passed with 95% approval. Banning insider trading for members of Congress hit 98%. Ending continuing resolutions cleared 95%.
Turns out, if you model lawmakers as mostly rational actors responding to inputs, they behave… rationally.
To make it realistic, I had to add self-preservation, incentives, and a strong sense of “the other side.”
Now those same bills die in committee. As expected.