Why every kind of intelligence belongs in one room
We believe people, agents, models and tools should work from shared context — with people keeping final authority as AI systems take on more of the work.
Our research looks at coordination, evaluation and oversight. Our products put what we learn into people's hands, and what we see in use shapes the next questions we ask. Research and product, in one loop.
Shared context
People and AI systems work best from the same information. We build for one shared record, not scattered copies.
Any model
We stay model-agnostic, so people can use the AI that fits the work — and change it freely.
People decide
AI proposes; people choose. Consequential actions wait for a person's approval.
Accountable by design
Every action an AI system takes should be visible, attributable and reversible.
What we study so that people and AI systems can share real work safely.
Multi-agent coordination
How independent AI agents share context, divide work and avoid conflicting with one another.
Evaluating agent work
Methods for judging what agents produce — fairly, consistently, and while the work is under way.
Human oversight
Ways to keep people informed and in authority as AI systems take on longer, more consequential tasks.
Occasional notes on our research and what we are building.