Refunds
The right amount, once, under the right policy.
A safer place for agents to act
Test action-taking support agents without touching production.
InvinciAI is exploring disposable, production-faithful Shopify and Gorgias environments for teams evaluating AI support agents.
Working on this problem? Email me.A disposable world
for consequential actions.
- 01
Create a clean world
Known orders, customers, tickets, policies, and edge cases.
- 02
Run your existing agent
Point it at Shopify- and Gorgias-compatible endpoints.
- 03
Inspect what changed
Trace actions, resulting state, and unintended effects.
Reset and repeat safely
The places where “mostly right” is still wrong.
Support agents do more than draft replies. The test environment has to remember what they changed.
Cancellations
Before fulfillment moves—and with every downstream change visible.
Address changes
Against the correct customer, order, and fulfillment state.
Reships
With inventory, fulfillment, and the support record kept in sync.
Ticket updates
Tags, notes, status, handoffs, and customer-visible replies.
State, not stubs.
An action should change the world the next action sees—across orders, tickets, customers, and fulfillment.
Outcomes, not demos.
Evaluation should reveal the final state and the path taken, not just whether an API returned 200.
Disposable by default.
Every run begins from known conditions, produces inspectable evidence, and resets without cleanup.
Research and early design-partner conversations.
We’re studying how teams test these workflows today, where existing mocks fail, and what a useful first environment must reproduce.
If your agent already takes actions in Shopify or Gorgias, we’d like to compare notes.