Our first research documents

We have published the first two documents in our Research section. One is a Paper, a more complete piece of work; the other is a Note, a work in progress we expect to revise.

The first is on organizational memory. AI agents are replaced each time the underlying models improve. For work to carry on regardless, what agents and people learn on the job has to be kept as knowledge that any agent can pick up, held by the organization rather than by any one model. The note sets out how we think about this.

The second is on running an AI-native company. When AI agents do most of a company’s work, it helps to keep three things apart: the rules and skills that define the agents, the data they work on, and the machines they run on. With these kept separate, agents can be added or removed as the work requires. The note describes the structure we have started to use in our own operations.

We are testing both ideas in our daily work and in Moss, our ongoing research and development project. We will revise the notes as we learn more, and we welcome comments from readers.