Operating System for Enterprise AI Agents

I’m building Braid OS for a future where organizations operate thousands or millions of agents. At that scale, the interface cannot be a longer list of agents. It has to tell the right person what matters now.

Agent activity is not the same as intelligence. A leader needs to know where execution differs from the plan, why it happened, and whether to intervene—without asking the right question first or reading every run.

Braid is designed to connect organizational intent to evidence from agent work. Its operating view surfaces the delta between expectation and reality in context: a finance leader and an engineering leader should see different answers from the same underlying activity. Agent-graph telemetry provides the trace; the product’s job is to turn it into a useful decision.

Braid is a working, invite-only system I’m dogfooding. Its agent-graph telemetry already makes Honey and Specular runs inspectable and governs agent packs in partner repos; I’m building on that foundation to surface the operational changes and decisions that matter to each person.