Docs / Operate & review
OPERATE & REVIEW / AI POD

How execution works

Dependencies, context checks, worker leases and final assembly.

Updated October 9, 2026 · 4 sections

Plan and queue

The Orchestrator creates dependent tasks. Starting a Pod queues work for the background worker. The worker runs eligible tasks only when their dependencies and required human gates are satisfied.

Context and evidence

Each job checks the pinned company version and selected evidence. A provider receives the role instructions, a scoped context object, upstream task outputs and any revision feedback. The Workforce Orchestrator checks acceptance criteria, sources, rules and assumptions before final consolidation.

Pausing and interruption

Pause or Stop all pods prevents in-flight output from being saved after the control changes. A provider request already sent can still incur charges. An expired worker lease becomes blocked after two minutes and can be explicitly retried.

Final result

The result keeps a review status. A person should inspect it before using it externally. The current worker drafts text; it does not verify a manual external action or execute publishing, payment or deployment.