Use case
Simulation agents that run while you sleep.
Spin up a full stack, point agents at real user flows, and get pass/fail reports without keeping a local session open.
Why this is hard
Unit tests pass and checkout still breaks
The failures that reach users live between services: a session that expires, a webhook that never fires, a form that submits twice. Nothing that mocks its dependencies will find them.
Manual QA does not scale with release cadence
Someone clicking through onboarding before every deploy is the first thing dropped when a release is urgent, which is exactly when you need it.
A failing test tells you less than watching it fail
An assertion error names a line. It does not show you the modal that covered the button, which is usually the actual bug.
How a run works
The stack comes up
Postgres, your API, the web app, and a real browser start together on one computer, seeded with data that looks like production.
The agent is given a flow, not a script
"Buy a subscription with a declined card, then retry with a good one." It works out the clicks rather than replaying a recording that breaks when a selector moves.
It drives the product like a user
Real browser, real network calls, real emails landing in a test inbox. If a human could do it, the agent can attempt it.
It reports pass or fail with the recording
You get the verdict plus a video of the session, so triage starts from what happened instead of a guess about what might have.
What you get back
- Pass/fail per flow
- Browser recording
- Service logs
- Screenshots on failure
What makes it real
Postgres, API, web app, and browser on one machine
Mail catcher and queues run beside them so a checkout flow can fail on a webhook that never fired, not on a mocked dependency.
Snapshots with the users and history a flow needs
Restore production-shaped data instead of rebuilding fixtures before every run.
Recordings and logs shipped with the verdict
When something breaks, triage starts from the session video and service output — not a stack trace line.
Who runs this
Islo proved for me that the dream of having an actual software factory is actually realistic.
Where teams go next
Common questions
What can a simulation agent test?
They run end-to-end flows such as checkout, onboarding, and regression suites against a live stack with browser, API, and database services.
Does a simulation agent need my whole stack running?
Yes. Islo restores the repo, database, API, web app, browser, and supporting services on one computer so the agent tests the product as a connected system.
Can a simulation agent test login, email, and seeded-data flows?
Yes. Run the services the flow depends on inside the development environment and give the agent scoped access to test accounts, inboxes, and seeded data.
How do I know whether the simulated user flow passed?
The agent returns pass/fail results and evidence from the browser and running services, so you can inspect what happened without replaying the flow locally.