Agentic SOPs

Written like an SOP. Run by an AI agent.

An agentic SOP is a procedure written for an AI agent to execute: when it applies, the steps and how it ends. In Aide you write them in plain English, test them on real conversations and see every run.

How do you write an SOP an AI agent can follow?

Write it as an agentic SOP: when it applies, the steps in order and how it ends, in plain English an AI agent can execute. In Aide these are procedures, which you test on real conversations before setting them live, and every conversation shows which one the agent followed.Updated
What an agentic SOP checks and does
When to useStepsCustom connector actionsActions governedChecked and testedEvery run on recordFix one procedureUse when examples
Only if conditionsTicket and contact fieldsSalesforce contact fieldsIf and OtherwiseHand-off rulesVersion historyRuns shown per message
76%+
AI resolution rate
The share of conversations resolved end to end, on paths your team pre-approved
100%
Accurate on what it acts on
Because it acts only within the rules and confidence your team sets, and hands off otherwise
24h
To first value
Topics from your own conversations, a first procedure, tested and set live
We currently use Zendesk and as such needed an AI solution that works with our existing CX platform and allows us to fully leverage our Shopify data to cut down support times and costs. Aide was a perfect fit as it integrates with both and seamlessly helped our agents get through tickets faster and automatically populates ticket fields on Zendesk!
Patchstore24Shopify
When students get complete, accurate answers to their questions quickly via email, they don't need to call or visit in person for clarification. The second-order effect is that our phone volume decreased and walk-in inquiries became less frequent.
Kristy Wachs from Stay UprightCustomer story
Really great approach and strong answers out of the box. Ability to grow into your AI journey at your own pace starting with suggested responses, macro suggestions or precanned messages and then slowly automate once you're confident.
Zendesk Marketplace reviewZendesk
Anatomy of an agentic SOP

When it applies. What it does. On the record.

A standard SOP was written for a person, and it relies on judgment to fill the gaps. An agentic SOP is written for an AI agent to execute, so every part is explicit. Here is each one.

Each one starts from a plain-English When to use, including what it is not for: "A customer disputes a card charge. Not a question about a pending payment." Each customer message follows at most one procedure.

Steps are explicit, never one block of prose a model interprets loosely. A course-reschedule SOP reads the Arlo registration, collects the new date with @, then hands off to your training team with a summary.

A step runs an action in your own systems through a custom connector, with the same confirmation, approval and log as an order action. Seats requested by email and approved by a teammate are added through your Billing API connector.

No action runs on an SOP's word alone. A refund opens only once the shopper is verified, follows the topic's AI level and the customer's confirmation, and waits for a teammate by default.

Check trigger tries the When to use on your last 50 conversations. Test replays a past conversation with the draft and sends nothing. Edits wait as a draft until an owner or admin sets them live.

Each customer message names the procedure Aide followed and the ones it considered, with the reason. Version history records who set each version live, when, and with what note.

With one monolithic instruction set, a single bad reply makes the whole set suspect. With agentic SOPs, you see which one Aide followed, change that procedure and test it again, while every other one keeps running untouched.
Aide is the AI-first customer service platform. In Aide, agentic SOPs are built as Procedures: a plain-English When to use and steps Aide follows, within the AI level of each topic. See how they are tested on real conversations and how Aide improves every night.
The Agent Governance Engine

Governance built into every procedure

An agentic SOP decides what the agent does when it applies. Your team's settings decide how far it goes.
Tested, then rolled outEach inbox moves from Watch only to test senders, then a share of conversations, then everyone.
Within the topic's AI levelEvery topic is at AI Agent, Drafts or None, and new topics start at Drafts.
On the recordEvery change is in version history and the audit log, with who made it and what changed.
Better every nightAide learns from your team's replies every night, tests each change before keeping it, and undoes it if quality drops.

Frequently asked questions

Agentic SOPs are standard operating procedures written for an AI agent to execute. Each one says when it applies, in plain English, and the steps the agent takes. In Aide they are called procedures: you check each one on your last 50 conversations before it goes live, and every run shows in the conversation.

A regular SOP is written for a person: prose steps that rely on judgment to interpret. An agentic SOP is written for an AI agent to execute, with a When to use the agent learns from examples, Only if conditions checked against live data, and explicit steps. In Aide the agent follows it one step at a time, and anything past the topic's AI level goes to a person.

A monolithic instruction set describes agent behavior in aggregate, so no single behavior is checked on its own and one bad reply makes the whole set suspect. Agentic SOPs cut the problem into procedures. In Aide each one is tested on its own, named under the message it handled, and fixed without touching the others.

Procedures is the product, and agentic SOPs is the idea behind it. Each Aide procedure is one agentic SOP: a plain-English When to use and steps Aide follows, tested on your real conversations and recorded in the Action Trace. See aide.app/platform/procedures.

In Aide, a procedure's steps take helpdesk actions: add a tag, fill ticket fields, assign a group, run a macro. Order actions on Shopify open once the shopper is verified and follow the topic's AI level and the customer's confirmation, and refunds wait for a teammate by default.

Aide connects your own data sources and internal systems through an API or an MCP server, so a procedure looks things up and takes action in them. The action runs for a customer Aide can verify, such as the requester on an email, with the same customer confirmation, teammate approval and log as an order action.

In Aide, Check trigger tries a procedure's When to use on your last 50 conversations and gives a reason for each near miss, and Test replays a past conversation with the draft without sending anything. Edits stay a draft until an owner or admin sets them live.

Each customer message in Aide names the procedure Aide followed and the ones it considered, with the reason, and the topic chip shows how confident Aide was. Version history and the audit log record every change to a procedure, with who made it.

Most teams write a procedure for their highest-volume topic first. In Aide a person sends each reply until you raise that topic from Drafts to AI Agent, and you change what runs on its own at any time.

Aide plans start at $39 a month and every plan starts with a 7-day free trial. You pay for AI replies, not seats. See aide.app/pricing for every plan.

Write the procedure once. Govern it forever.

Procedures your team checks on real conversations, sets live when ready, and sees on every run.

Not ready yet? Get product updates by email