Federation degrades continuously and invisibly. A credential rotates, a source starts timing out, a network path gets slower, and nothing dramatic happens — queries just get worse until an analyst complains that a dashboard looks wrong. The failure mode of a federated platform is rarely a clean outage; it is a slow drift nobody is watching.
SemanticFed already records the signal. Every connection probe against a data source captures a health verdict, a latency measurement, and a message, and each data source carries a connectivity lifecycle with validated transitions between its draft, connected, degraded, disconnected, and disabled states. What has been missing is not data. It is action: a human reading the probe history and driving the transition by hand.
The autonomous loop closes that gap. On a cadence, and reacting to new probe events as they arrive, it reads probe history per source, decides whether the observed reality justifies a lifecycle transition, applies it, and notifies the source's owner. Because it drives the same validated transition path a human uses, an illegal transition proposed by the agent is rejected by the existing validator rather than special-cased for the machine.
This is the one loop in the design permitted to act rather than only propose, and the reasoning matters: reconciling a lifecycle state to observed health is reversible and merely records reality. Widening or narrowing data access is not, which is why the policy loops in this product propose only. Every autonomous action is designed to be recorded with its trigger and decision in a shared agent-run ledger, metered against the workspace's AI budget, and stoppable per workspace with a single flag — an autonomous agent without an off switch is a defect, not a feature.
The loop works on the same probe records, lifecycle state machine, and transition validator a person uses, so it can never take a step a person could not.
Illustrative scenario — not a record of a real deployment or a customer result.
Ready to make this your story?
