“If it must survive a restart, it is a row” is the line I’m stealing. The visibility point matters even more with agents than with normal jobs: a queued session that looks stuck tends to get re-dispatched by an impatient user, and then you pay for the same tokens twice. (Disclosure: I’m with aicost.tools, so double-billed agent runs are my pet peeve.) Do you dedupe dispatches on the Postgres side, e.g. one queued session per task, or did the UI fix make that unnecessary?
1 Like