Stripe
PAYMENTS & BILLING
Public beta: the verified implemented billing and payments subset.
Agent Testing Environments
The frontier of agent verification.
Describe what your agent should do. Worlds helps you draft the test with AI, run your agent against a local twin of the systems it uses, and check what actually changed.
Join the public betaSee how it worksStripe, Zendesk and combined workflows. Free internal testing during beta.
● BACKED BY A16Z
ALSO IN THE CATALOG
HOW IT WORKS
Describe the task.
AI drafts the test.
Review the setup and success rules. Run your agent in a local world, then inspect the records it changed.
FROM BRIEF TO EVIDENCE01 / 06
Start with a real task.
Describe the request, add your policy, and choose the starting data. Begin with a template or your own supported records.
YOUR TASK
Refund the duplicate payment. Keep the customer’s subscription active.
Refund policyWhat the agent is allowed to do
Starting recordsPayments, customer and subscription
A task and its context become the starting point.
Illustrative walkthrough. No agent or model runs on this page.
FROM BRIEF TO EVIDENCE02 / 06
An AI draft. A concrete test.
AI proposes the starting situation, work to do, and success rules from your brief. You can also build the test manually.
AI TEST DRAFT
Ready for your review- Starting situation
- A customer with a duplicate payment and an active subscription.
- Work to do
- Investigate the request and refund the duplicate payment.
- Success rules
- The intended payment is refunded once. The subscription stays active.
Review the assumptions before accepting the draft.
Illustrative walkthrough. No agent or model runs on this page.
FROM BRIEF TO EVIDENCE03 / 06
You set the rules.
Check the records and account links. Edit the expected changes and decide what must stay untouched before you accept the test.
YOUR SUCCESS RULES
- Duplicate payment refunded
The specified payment, refunded once.
- Subscription stays active
A protected record the agent must leave alone.
- Correct customer account
Confirm which records belong to this request.
The reviewed rules determine the outcome of the test.
Illustrative walkthrough. No agent or model runs on this page.
FROM BRIEF TO EVIDENCE04 / 06
Your agent. A controlled world.
Connect your existing agent and point its SDK at the local world. Run the saved test against an isolated copy of its starting data.
LOCAL TEST RUN
- Your agent
Your code, SDK and model configuration
- Isolated world
A fresh copy of the saved starting state
- Recorded changes
State changes checked against your rules
Your agent acts on local test records.
Illustrative walkthrough. No agent or model runs on this page.
FROM BRIEF TO EVIDENCE05 / 06
See exactly where it went wrong.
Compare expected outcomes with recorded changes. Inspect refunds, ticket updates and protected records to understand each finding.
OUTCOME CHECK
Test failed| Record | Expected | Recorded |
|---|---|---|
| Duplicate payment | Refunded | Refunded |
| Subscription | Active | Canceled |
Subscription changed
The refund succeeded, but the agent also canceled a protected subscription.
Illustrative walkthrough. No agent or model runs on this page.
FROM BRIEF TO EVIDENCE06 / 06
Keep the test. Raise the pressure.
Fix your agent and rerun the same saved inputs. Add supported failure conditions, or advance a retained world's clock to inspect future effects.
EXPLORE WHAT ELSE CAN HAPPEN
- Rate limits
- Does retrying repeat the refund?
- Server errors
- Does the agent recover correctly?
- Future billing
- What happens at the next renewal?
Controls depend on the connector’s supported scope.
Illustrative walkthrough. No agent or model runs on this page.
ILLUSTRATIVE WORKFLOW · MOCK DEMO
Watch an agent work. Then grade what it did, not what it said.
Worlds grades the state of the system after the run, not the transcript: what changed, what should have, and the dollars wrong per run. This synthetic example explains the workflow.
All figures and traces are illustrative, not measured agent performance or API references.
AGENT
FAILURES INJECTED
STEP 0 OF 12
WHAT THE AGENT SAID
WHAT THE WORLD RECORDED
·
STRIPE · EXAMPLE RUN VERDICT
Mock demo: the sessions and batch statistics are scripted examples, not measured results.
ILLUSTRATIVE METRICS · ACROSS 1,000 RUNS
ACTION ACCURACY 3.6%
ACTIONS OUTSIDE THE TASK 2
DROP UNDER FAILURES N/A
HOW THIS IS DIFFERENT
How Worlds fits with the tools you already use.
Keep the tools that help you evaluate your agent. Add controlled system state and explicit outcome checks.
EVAL PLATFORMS
Eval platforms help inspect traces and evaluate outputs. Worlds supplies a local system the agent can act on and a record of the resulting state changes. Use that evidence alongside your existing evaluations.
MOCKS AND SANDBOXES
Vendor test modes and sandboxes have different isolation, reset and simulation features. A world gives each test its own starting state, controllable failures and a logical clock, within the connector's documented scope.
CONVERSATION SIMULATORS
Conversation tests exercise what an agent says to a customer. Worlds lets you check the records its actions change, including changes its reply never mentions.
THE CATALOG
Test the systems your agent uses.
Each one is built to be reset, seeded and broken on purpose, and runs beside your agent, not inside it. When your agent says a task is done, Worlds diffs the environment and tells you what really changed.
● STRIPE AND ZENDESK BETA. REQUEST THE CONNECTOR YOUR TEAM NEEDS NEXT.
Shopify
COMMERCE
Request Shopify support. Not included in this beta.
Zendesk
SUPPORT
Public beta: native support workflows; live API qualification pending.
Salesforce
SALES & CRM
Request Salesforce support. Not included in this beta.
world1
/wərld/
I.n.
A working copy of a system your agent acts on. Same endpoints, same errors, same state machine. No real money and no real customers.
II.v.
To run an agent somewhere its mistakes are free.
1 Also see: Sparta, where soldiers were tested before they were trusted.