AI agent operations: how AR Labs runs

AR Labs is run as a small company operated by one person, Richard M. Thompson, who founded it, and a set of AI agents (language models run in a loop with tools, so they can act, not only reply). The company structure is deliberate: it lets us test, at the scale of one person, how agents take on whole roles while a person keeps the final say. Work is divided into departments, a strategy agent runs planning cadences (recurring planning reviews, from yearly to daily), agents hand work to each other through a shared file mailbox, and dashboards and maps show where everything stands.

The pages in this section describe that method and the tools behind it, for anyone running AI agents against a plan: how strategy flows down and status flows up, how departments and activity streams divide the work, how escalations reach a decision, how agents hand work to each other, how the repository acts as long-term memory for agents that start every session with an empty context window, how session history is kept, what an operations dashboard shows, and how a week is scheduled against the plan. The skills that run these methods are described under Claude Code skills.

The pages describe structure and mechanism only; our actual goals, figures and priorities are not published. Where a part of the system is a design rather than something running today, the page says so.