Should the author gradetheir own work?
Keep using your coding agent to build and to check its work as it goes. Smoketest takes over when the ticket reaches QA. It did not write the code, it starts from what the ticket asked for, and it leaves the proof where your team reviews it.
Failed · 1 of 4 requirements not met · 2m 38s
The invoice PDF leaves out the proration line.
Watch the recording · View attempt
Your coding agent tests what it built.Smoketest tests what the ticket asked for.
The agent says done.Who checked?
Here is one ticket, built by a coding agent. Its tests pass, because it wrote them for the code it wrote. The ticket also asked for the PDF.
Show proration on the invoice PDF
When an Admin adds a seat mid-cycle, the invoice PDF should carry the proration line and match the invoice preview.
Done. Added the proration line to the invoice preview.
All 42 tests pass, including the 3 added for this change.
Failed · 1 of 4 requirements not met · 2m 38s · Attempt 1
The invoice PDF leaves out the proration line. The preview says $312.40 and the PDF says $294.00.
Next: moved to In Progress. Move the ticket back into QA to test again.
Watch the recording · View attempt
The same eight questions, asked of both.
Most teams need both.
A coding agent is the right tool while you build. Here is when a separate check earns its place.
Your coding agent is enough when
- One person builds, reviews, and ships the change
- A bug in production costs you little and is quick to fix
- You are checking a change while you work on it
Add Smoketest when
- Agents write more code than your team can review
- Someone else needs proof that the ticket really works
- The flow takes a login, an email code, or a PDF to test
It did not write the code.
That is the whole point. A second agent with a different starting point has no reason to go easy on the change, and every step it takes is on the recording for a person to check.
Keep your agent. Give it a reviewer.
Nothing about how you build has to change. The ticket gets a check from someone who was not in the session.
Your agent builds
It writes the change and checks its own work as it goes, the way it does today.
The ticket moves to QA
Smoketest reads the ticket and tests what it asked for, in a real browser or on a device.
Your agent fixes
A failed result carries the step and the recording, and Claude Code or Cursor can read it over MCP.
What it takes to build this yourself
- Test accounts and saved sign-ins for each role
- An inbox for email codes and magic links
- Recordings a person can review
- A way to post results back to Jira or Linear
- Someone to look after all of it next quarter
Our founder sets it up with you on a real ticket.
A call with our founder
Tell us how your team ships and where tickets wait. We will tell you whether it fits.
A real ticket, same day
We set it up with you on your own board, so it can start the day of the call.
A plan shaped to you
It starts from $399 a month. Scale is designed with teams that need more runs.
It changes as you do
New flows, new tools, a new QA hire. We keep adjusting it with you as you grow.
Questions engineers ask first
Seeing this on your own team? Read how teams check AI-generated code
Claude Code is good for a quick check while you build. The catch is that the agent that wrote the change is then grading its own work. Smoketest starts from the ticket, did not write the code, and leaves the proof where your team reviews it.
It can be wrong, so it shows its work. Every result has the recording and the steps, and says what it did not test. When it cannot tell, it comes back as No Result and asks a person.
Yes, and some teams do. The first run is the easy part. The work is in test accounts, sign-ins, email codes, recordings, and posting results to the ticket, and in keeping all of that running.
Yes. Smoketest has an MCP server, a CLI, and an API, so your coding agent can start a test, read a failure, and fix the code with the recording in hand.
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.
A named person on your team. Smoketest gives them the evidence: what passed, what failed, the recording, and what it did not test. It never signs off a release.
Let your agent build. Give us the ticket.
Book a call and we will test one ticket your agent finished. You get the result and the recording the same day.