The agent says done. Who checked?
AI coding agents write code faster than a team can review it, and they grade their own work. Smoketest tests each ticket from the outside, so you can see whether the change does what you asked before it ships.
Done. CSV export added, and all tests pass.
Failed · 1 of 4 requirements not met · 3m 12s · Attempt 1
- ✓Admin can open Usage
- ✓Export downloads a CSV file
- ✓CSV columns match the ticket
- ✗CSV uses the selected date range
Not tested: exports over 100,000 rows, because the ticket does not mention them.
Watch the recording · View attempt
Good catch. Sending this back to the agent.
It tests what you asked for.
Does it do what the ticket says?
It starts from the ticket, so it checks the request as well as the code.
What did the change touch nearby?
It reads the pull request and tests the flows around the change.
Does it work for a real user?
It signs in with your test accounts and uses the product the way they would.
What usually slips through
These come up on tickets written by people too. They turn up more often when code arrives faster than anyone can read it.
What the ticket left out
Empty states, other roles, and bad input are easy for an agent to skip.
The screen next door
It also changes another screen that shares the code.
Tests that match the code
The agent wrote the tests too, so they check what it built.
One line of the ticket
It works, and still misses one thing you asked for.
Questions about testing AI-generated code
It can be, once something other than the agent has tested it. That check matters more as code arrives faster than your team can read it.
Keep those tests. They check what the agent built. Smoketest starts from the ticket, so it checks what you asked for, and the proof lands where your team reviews work.
No. It tests the running product the way a user would. Code review stays with your team and the tools you use for it.
Yes. They can start a test and read the result over MCP, so an agent can fix what failed before a person looks at it.
It starts at $399 a month for 1,000 runs. Each run is up to 10 minutes, and a longer test counts as two runs or more, one for every 10 minutes. If your team needs more, we design a Scale plan with you, so you stop buying extra runs.
Every result links to the recording, the steps, and the reasoning, and says what it did not test, so you can check its work the way you would check a teammate's.
Give it a ticket your agent finished.
Book a call and we will test one with you. You get the result and the recording the same day.