How should an AI pipeline represent job states?

How should an AI pipeline represent job states?

Job states describe the current position of a generation request. They should be stored consistently so the interface can recover an accurate view after a reload.

Define states and transitions

A simple design might use queued, running, awaiting review, completed, failed, and cancelled. These are example names; select the states your workflow actually needs. Define which actor can move a job between states and which conditions must be true. Completion should require verified output, not merely acceptance by an external service.

Handle late and repeated updates

A worker may finish after cancellation or retry an earlier update. Include an attempt or version identifier so stale work cannot overwrite a newer result. Decide how terminal states behave and whether restarting creates a new attempt. Walk through duplicate completion messages and a user cancellation during rendering. The goal is a state model that preserves the real history instead of making the UI appear successful prematurely.