Edit an order or swap an item with an AI agent
Wrong size, wrong color, one more item. The AI edits the Shopify order before it ships, so the customer does not return it later.
Last updated September 24, 2026
~3 min read
By Amit RG, Founder
A swap instead of a return
A wrong size caught before shipping is one edit. Caught after delivery, it is a return label, a second shipment and a refund or exchange.
For subscriptions, the same swap happens in Recharge. See subscription actions.
Who approves an order edit
Edit Order has its own permission: Always allow, or Require approval. When it waits for approval, the edit itself is the review card your agent approves in one click.
An action with no permission set waits for approval by default. Any refund the edit creates follows your refund permission separately.
When the AI hands the edit to your team
Shipped orders go to your team. So do items the AI cannot match and edits your policy does not cover.
A hand-off takes three steps. The AI saves an internal note with the reason. It sends the customer one reply on what happens next. The conversation stays open for your team.
If your playbook names a hand-off mode, the AI switches the conversation to Manual or Co-Pilot. No further AI reply then goes out on its own.
Which systems the edit runs in
| System | Edit action |
| Shopify | Order editing: swap, add, remove, change quantity |
| Shopify | Draft order editing |
| Shopify | Live shipping rates for a shipping-method change |
| Shopify | Product search for the replacement variant |
| Shopify | Invoice for any balance the customer owes |
Every other AI action, and how each is controlled, is on the AI actions overview.
Where these facts come from.
- Richpanel AI action catalog, as of September 24, 2026. The list of actions the AI agent can call, with each action's inputs, preconditions and default permission. Every capability on this page maps to a named action in it: edit_order, edit_draft_order, check_shipping_options, search_products, get_product_details and send_invoice.
- Richpanel AI safety. The five safety layers, including sensitive-action guardrails, tool permissions and QA AI review of every conversation. richpanel.com/ai-safety
Version history. v1.0 (2026-09-24): initial publication. Every action on this page is checked against the Richpanel AI action catalog as it stood on the publication date.