AI agents

AI agents answer customer questions automatically and escalate what they shouldn't handle. They're available on all paid plans, charged per successful resolution.

What agents can do

• Answer questions from your knowledge base — shipping, returns, sizing, policies.
• Look up real Shopify data — order status, product details, stock.
• Recommend products.
• Collect information from customers, including WhatsApp consent.
• Add someone to a back-in-stock waitlist.
• Hand over to a human when they should.

Agent purposes

You can run more than one agent — a sales agent on webchat where visitors are browsing, and a support agent on WhatsApp where existing customers are asking about orders.

Configuration

Writing instructions

Instructions govern behaviour. Facts belong in the knowledge base — putting them in instructions means they're sent on every request whether relevant or not.

Good instructions are specific and mostly negative:

Be concise — two or three sentences unless asked for detail.
Never promise a refund, discount, or delivery date. Hand over instead.
If the customer seems frustrated, hand over rather than trying again.
If you don't know, say so and hand over. Never guess at policy.
Always hand over if the customer asks for a person.

Vague instructions ("be helpful and friendly") don't constrain anything. The valuable ones tell the agent what not to do.

Shopify tools

Agents can look up live store data through tenant-scoped backend tools — order status, product details, stock levels, and discounts where you allow it.

This is what separates a useful agent from a chatbot: "where's my order?" gets a real answer with a real tracking status, not a link to a tracking page.

Tool calls run server-side and are scoped to your store. An agent cannot reach another merchant's data.

Where to enable agents

Start with webchat. Storefront visitors ask the same pre-sale questions constantly — sizing, shipping times, stock — and those are highly automatable with low risk.

WhatsApp is more valuable but higher stakes: it's usually existing customers with order problems, where a wrong answer costs more.

Facebook and Instagram depend on your volume. A low-traffic Page may not justify it.

Testing before you go live

Ask the questions your customers actually ask, including the awkward ones:

• Your top five real support questions.
• The exceptions — "can I return a sale item?"
• Something you deliberately haven't documented. The agent should hand over, not invent an answer.
• Anything about money. That should escalate.
• "I want to talk to a human." That must hand over immediately.

The last two are the ones worth being strict about.

Resolutions and billing

You're charged only when the AI solved the problem and closed the conversation with no human involvement.

Because escalation is free, tune your agents towards handing over readily. An agent that escalates too often costs you nothing extra; one that guesses at a refund policy costs you a customer.

Rates are $0.18 to $0.12 per resolution depending on plan, plus $0.032 when audio or images were involved. See AI usage and spend.

Agents and workflows

AI agents and workflows do different jobs.

Workflows are deterministic — the same trigger always produces the same message. Use them for anything predictable: order confirmations, shipping updates, review requests. They cost nothing per run beyond your allowance.

Agents handle the unpredictable — open-ended questions you can't anticipate.

Don't use an agent for something a workflow handles. An order-status notification triggered by a Shopify event is free and always correct; an agent answering "where's my order?" costs a resolution and might get it wrong.

Two connection points between them:

• The Assign to Agent node routes a conversation to an AI agent mid-flow.
• The AI Handed to Human trigger starts a workflow when an agent escalates.

Troubleshooting

Next steps

• Knowledge base — what agents answer from.
• Handover — escalation to humans.
• AI usage and spend — cost control.