Bring a failure
we can investigate.

Or a better way to test what helps. We welcome researchers and engineers working on agent behavior, memory, evaluations and governance.

A useful first conversation.

We are developing an automated research loop for agent-swarm governance. It will investigate failures, propose safeguards and test whether they help. Our first application asks how a group recovers after a bad method spreads beyond the copies we can find.

You might have seen a related problem, spotted a missing comparison, or built a memory or evaluation system that should be part of the test. We would like to understand what you found.

A stronger study
Challenge our assumptions, propose a baseline, or help distinguish an intervention’s effect from another explanation.
Better research tools
Help connect messages, saved knowledge and later actions, or test whether automated investigation makes the research more reliable and less costly.
An independent test
Repeat a comparison or try it in a different environment as the tools and findings become available.
Read the research approach

The first study is in preparation. We are seeking critique and research collaboration; this page does not list open employment positions.