MyTest is an AI QA platform in which coding agents test your app and a second agent checks their work. You describe your app and its key user journeys in plain English; MyTest creates testing agents, runs smoke tests against staging and saves the scenarios for future runs. A separate verifier agent re-runs each failure and only reports it as a bug if it reproduces. It was created by Sakthi Vignesh and is live at mytestagent.vercel.app.
What is the problem with AI testing agents?
Getting an agent to click through an app is the easy part. The hard part is trust. A testing agent that reports a failure that is not real, because the page was slow, the agent misread a button or the data changed, puts noise on the bug board. After a few false alarms, developers stop reading the reports, and the tool has made testing worse.
How does the verifier agent work?
The roles are separate on purpose. One set of agents runs the scenarios and records what went wrong. A second, independent verifier agent takes each reported failure and tries to reproduce it. If it reproduces, it becomes a bug report; if it does not, it is dropped or flagged as flaky. The principle is the same as asking a colleague "can you see this too?" before filing a ticket.
What evidence does a failure include?
- Screenshots
- Console logs
- Network traces
A bug report with evidence can be triaged in a minute; one without it starts an argument.
What does it work with?
MyTest works with Claude Code, Claude Desktop, Antigravity and PagePilot. It is built with AI agents, Claude, MCP, Python and TypeScript.
Vantaverse's Take
The verifier is an application of a rule we use on every agent project: never let the thing that produced a result be the only thing that judges it. The same idea shows up in how we test client agents before launch, which we describe in how we evaluate AI agents and the longer research note on reliability. See the MyTest product page for more.