Agent operating procedure
A written procedure that tells an AI agent how to handle a type of work, encoded from the standard operating procedures a finance team already follows.
Every AR team has procedures: when to escalate, which customers get a call instead of an email, what documentation a dispute needs. Many live in people's heads or in documents nobody has updated.
Writing them as agent operating procedures makes an agent's behavior explicit and reviewable. When policy changes, the procedure changes, and the agent follows the new version.
Related terms
- Supervised autonomyGiving an AI agent independence in stages: it starts by proposing actions for approval and runs on its own only as its work proves reliable.
- BacktestingTesting a change to an automated process against past cases before using it on live work, to see what it would have done differently.
- Human in the loopA design in which an automated system hands defined decisions to a person to review or approve before it acts.