An agent operating procedure (AOP) is a natural-language instruction set that tells an AI customer service agent how to behave across the situations it encounters. Agent operating procedures give teams a flexible, plain-language way to encode company policy and process for an AI agent.
An AOP is powerful because it is expressive. You write instructions in natural language and the agent interprets them. The tradeoff is that an AOP is monolithic. It describes behavior in aggregate rather than binding each rule to a specific, classified customer intent, so it can be hard to test one slice in isolation or to measure which customer needs are actually covered.
Aide, the agentic AI platform for customer experience, offers the intent-scoped, governed evolution of this idea: ASOPs (Agentic SOPs). Where an AOP is one large instruction set, an ASOP deploys one intent at a time off the Customer Intent Map, is test-gated by the Agent Simulator against that intent's real conversation history, and is auditable per intent once live. The unit of work is the intent, not the whole policy document.
Scoping changes what a team can know. One broad instruction set cannot be tested a slice at a time; an intent-scoped procedure can. Per-intent coverage is per-intent legibility: the team can always say which customer needs run automated and which still sit with people, rather than watching its operation blur into one document.
Frequently asked questions
- What are agent operating procedures used for?
- Agent operating procedures are plain-language behavior specs for AI customer service agents. Teams use them to encode policy, tone, and process in one instruction set so the agent handles the situations it encounters consistently.
- What is the difference between an AOP and an ASOP?
- An AOP is a monolithic natural-language instruction set. An ASOP is intent-scoped, test-gated, and auditable per intent. The ASOP binds automation to a single classified intent rather than to aggregate behavior.