Patchstore24Shopify
Agentic SOPs
The next generation of SOPs,
built for AI agents to execute.
An Agentic SOP (ASOP) is one intent's procedure, written for an AI agent to execute: the conditions to check, the actions allowed, and the confidence gate to clear, tested on your real conversations before it goes live, and auditable per intent.What an Agentic SOP does on every conversation
One classified intentConditions, actions, gatesTest-gated before liveCoverage, measuredAuditable per intentOne intent isolated and fixedOne classified intentConditions, actions, gatesTest-gated before liveCoverage, measuredAuditable per intentOne intent isolated and fixedOne classified intentConditions, actions, gatesTest-gated before liveCoverage, measuredAuditable per intentOne intent isolated and fixed
Classify the intentCheck every conditionGate on confidenceTest on real conversationsLog every actionFix one intent at a timeClassify the intentCheck every conditionGate on confidenceTest on real conversationsLog every actionFix one intent at a timeClassify the intentCheck every conditionGate on confidenceTest on real conversationsLog every actionFix one intent at a time
76%+
AI Resolution Rate
The share of conversations resolved end to end, on ASOPs your team pre-approved
100%
On what it acts on, because governed
An ASOP fires only when every condition is met and confidence clears your bar
24h
To first value
Classify your queue, build your first ASOP, test it, deploy it
Trusted by customers on
Wicked ClothesShopify
Zendesk Marketplace reviewZendesk
Anatomy of an Agentic SOP
One intent. One procedure. Fully governed.
A standard SOP was written for a person to follow, and it relies on judgment to fill the gaps. An Agentic SOP is written for an AI agent to execute, so nothing is left to judgment: the intent, the conditions, the actions, and the gate are all explicit. Here is each part, shown in a real conversation.
An ASOP does not describe agent behavior in aggregate. It starts with scope: the Customer Intent Engine classifies the conversation first, and only that intent's ASOP can fire. Automation is bound to the intent, so no procedure ever reaches into a conversation it was not written for.
Customer
Hi, I sent back order #4092 last week and have not heard anything since. Where is my refund?
Aide
The carrier scanned your return for order #4092 on Monday, and your refund of $58.20 was issued this morning. It should reach your original payment method within 3 to 5 business days.
✓Intent classified: refund status · Only that intent's ASOP eligible · Fired
An ASOP is not prose a model interprets loosely. It is written in parts an agent can execute: conditions verified against live data, actions drawn from an explicit list, and a confidence gate the run has to clear. What executes is the procedure your team wrote, nothing beside it.
Customer
The sneakers in order #9316 run small. Can I swap them for a size 10?
Aide
Yes. Order #9316 is within the 30-day exchange window and the size 10 is in stock, so I have set up the exchange and emailed you a prepaid return label. The new pair ships as soon as the carrier scans your return.
✓Conditions met: window open · size in stock · Approved action: exchange · Confidence cleared the gate
No ASOP deploys on faith. Before it goes live, the Agent Simulator replays it against your real historical conversations for its intent, and your team watches exactly how it would have handled each one, adjusts it, and approves it. Know your AI works before your customers find out it doesn't.
Customer
I am traveling next month. Can you pause my coffee subscription for the next two shipments?
Aide
Done. Your subscription is paused for the next two shipments and will resume automatically on October 14. You will get a reminder email three days before it restarts.
✓Simulator replay · Subscription pause ASOP · 240 past conversations · Awaiting team approval
Because every ASOP maps to one intent, coverage is a number, not a feeling. Intent by intent, you know exactly what is automated and what still goes to a person. That is the Intent Coverage Rate: the depth behind the AI Resolution Rate you report.
Customer
Could you send me the invoice for order #3578? Finance needs it for our expense system.
Aide
Here is the invoice for order #3578, attached as a PDF and issued to the billing details on your account. Tell me if you need it made out to a different company name.
✓Invoice request ASOP · Live · 14 of 31 mapped intents automated
Every run of an ASOP is recorded and reviewable through the Action Trace: the intent detected, the conditions checked, the confidence at the moment of decision, and the action taken. You can audit one intent's automation without untangling everything else. Never wonder what the AI did or why.
Customer
Your WELCOME15 code would not apply at checkout, so I paid full price on order #6250. Can you fix that?
Aide
You are right, the code excludes sale items and one was in your cart. I have applied a 15 percent adjustment of $12.75 to order #6250 instead, so you end up exactly where the code would have put you.
✓On the record: intent, conditions, adjustment issued, confidence · Reviewable in the Action Trace
With one monolithic instruction set, a single bad reply makes the whole set suspect. With ASOPs, a failure isolates to one intent. You fix that procedure, retest it in the Agent Simulator, and redeploy it, while every other ASOP keeps running untouched.
Customer
The zipper on the backpack from order #8871 broke after a month. Is that covered?
Aide
It is. Order #8871 is inside the 12-month warranty, so I have approved a replacement. It ships within 2 business days, and you do not need to send the damaged one back.
✓Warranty ASOP updated for the new policy · Retested · Redeployed · Other intents untouched
Aide is the agentic AI platform for customer experience. In Aide, Agentic SOPs are built as Scenarios: one per intent, with the conditions, actions, and confidence gates enforced by the Agent Governance Engine.
Customer
Hi, I sent back order #4092 last week and have not heard anything since. Where is my refund?
Aide
The carrier scanned your return for order #4092 on Monday, and your refund of $58.20 was issued this morning. It should reach your original payment method within 3 to 5 business days.
✓Intent classified: refund status · Only that intent's ASOP eligible · Fired
The Agent Governance Engine
Governance built into every ASOP
Because every ASOP is scoped to one intent, governance is per intent too. The Agent Governance Engine enforces each procedure exactly as your team wrote and approved it.Tested per intentThe Agent Simulator replays each ASOP against your real past conversations for its intent. Nothing goes live until your team has watched it work and approved it.
Fires only when certainAn ASOP runs when every condition is met and confidence clears the gate your team set. Anything short of that goes to a person with context already attached.
On the record, per intentThe Action Trace records each run: the intent detected, the conditions checked, the action taken, and why. Auditable per ASOP, and across your operation.
The next ASOP, suggestedThe Continuous Learning Engine watches what your team still resolves by hand and suggests the next ASOP worth writing, with the evidence behind it.
Getting started
Works with your existing stack
An ASOP's conditions read live data where it lives, and its actions execute in the systems you already run. One click into your helpdesk, direct actions in your systems of record. No migration, no replatforming, no engineering.
Resolves and updates tickets inside your existing workflows.
Closes conversations end to end, right in the shared inbox.
Automates the macros and rules your team already relies on.
Reads live orders and executes changes, holds, and refunds.
Order data and actions, handled natively on your store.
Grounds every answer in the customer record you trust.
Frequently asked questions
Agentic SOPs (ASOPs) are the next generation of SOPs, built for AI agents to execute. Each one binds automation to a single classified customer intent and carries the conditions to check, the actions allowed, and the confidence gate to clear. An ASOP must pass a test on your real historical conversations before it goes live, and every action it takes stays auditable per intent.
A regular SOP is written for a person to follow: prose steps that rely on judgment to interpret. An Agentic SOP is written for an AI agent to execute: explicit conditions verified against live data, an explicit list of allowed actions, and a confidence gate. A person reads an SOP and decides. An agent runs an ASOP and either clears every gate or escalates to a person.
A monolithic instruction set describes agent behavior in aggregate. It is flexible, but no single behavior can be tested in isolation, coverage is not measurable, and when something breaks the whole instruction set is suspect. ASOPs cut the problem per intent: each one is tested on its own, coverage is measurable intent by intent, and a failure isolates to one procedure you can fix, retest, and redeploy. The difference is governance and scoping, not expressiveness.
Scenarios are how you build ASOPs in Aide. Agentic SOPs are the concept: intent-scoped, test-gated procedures written for AI agents to execute. A Scenario is one ASOP in the product, with conditions your team sets and actions your team allows, run under the Agent Governance Engine, tested in the Agent Simulator, and logged in the Action Trace.
It has to clear a test gate. The Agent Simulator replays the ASOP against your real historical conversations for its intent, so your team sees exactly how it would have handled each one: the conditions checked, the action it would have taken, and the reply it would have sent. No ASOP deploys before your team has watched it work and approved it.
Yes. The Action Trace records every run: the intent detected, the conditions checked, the confidence at the moment of the decision, and the action taken. Because every automation is scoped to one intent, you can audit one ASOP's runs without wading through everything else the system did.
One intent at a time. Start small. Prove it works. Expand from there. Most teams write an ASOP for their highest-volume intent first and see first value within 24 hours. Each new ASOP goes live only after it clears its test gate, and you can change what runs on its own at any time.
Get a demo for pricing.
Procedures and deployment, defined
The vocabulary of agentic procedures, defined.
- Agent operating procedure (AOP)
- Agentic SOP (ASOP)
- AI agent action
- AI agent workflow
- Business logic for AI agents
- CX workflow automation
- Decision tree vs agentic procedure
- Deterministic vs probabilistic agent execution
- Intent-scoped deployment
- Natural-language workflow authoring
- Resolution playbook
- SOP for AI agents
Write the procedure once. Govern it forever.
Intent-scoped automation your team tests before it ships, deploys one ASOP at a time, and audits on every action.