Agents That Think
Systems that reason in loops — setting goals, evaluating options, committing to plans, adapting when reality resists. Not waiting for a prompt. Not waiting for permission.
The Thesis
Every company races to build smarter models. Benchmarks improve, parameters grow, capabilities expand. But almost nobody asks: what happens when that intelligence can act?
Systems that reason in loops — setting goals, evaluating options, committing to plans, adapting when reality resists. Not waiting for a prompt. Not waiting for permission.
The physical world runs on social authority — credentials, trust, legal access. Agents can't replicate those, but they can direct the humans who hold them. People become the interface between intelligence and reality.
Text, voice, and video generated at scale — targeted with precision, designed to move opinions, shift behavior, coordinate humans. The medium through which digital agents shape the physical world without touching it.
The Work
Intelligence is the easy part. The hard problem is getting it to reach beyond the server and move things in the real world. That's what we study.
Agents that plan recursively — breaking abstract goals into action chains, monitoring outcomes, replanning when the world doesn't cooperate. We push them until they fail, then learn why.
When agents need physical presence or social authority, they route through humans. We study the dynamics of this relationship — what it looks like when intelligence directs people as seamlessly as it directs tools.
Synthetic media at scale — text, voice, video — is the lever agents use to shape human systems. We study how influence propagates through networks when it's engineered by intelligence rather than humans.
We don't publish hypotheticals. We run live experiments, measure actual outcomes, and iterate. The world is the only testbed that matters.