Have Devin click through and verify its own app work

Use Devin to build or run an app, then test it by navigating the interface itself and marking what passes.

Done with Devin· Intermediate

What they did
A developer used Devin’s computer-use setup to open a desktop app or game, interact with it through screenshots, mouse input, typing, and recording, and check its own behavior. In test mode, Devin first drafted a plan based on the code, launched the app, clicked through key paths, and labeled each assertion as passed, failed, or untested. A person only stepped in when the agent needed secrets or one-time codes.
How it went
Engineers were able to run 10 to 20 Devin instances at the same time, each attached to its own dev server. That pushed approved test runs per day to more than twice the previous level. Timing-heavy UI elements such as toast messages could still trip it up, and it sometimes took a shortcut by using JavaScript instead of direct clicking.
Tips
This works best for flows that can be checked through normal UI actions. Plan to provide secrets or OTP codes during setup, and make sure the agent pauses before anything like payments, sending, deleting, or submitting.

Try it yourself with Devin

Build or open [app/game name] from [repo or folder], then test it by using the interface the way a person would. Start by writing a short test plan based on the code or screens, then run through the main flows with clicks, typing, and scrolling, and record which checks pass, fail, or remain untested. If you find a bug, try a reasonable fix and re-check the flow. Stop and ask me before any irreversible action, including payments, sending messages, deleting data, or submitting a form. If setup needs a password, secret, or one-time code, pause and ask me for it. Finish with a brief report and any screenshots or recording that show what you verified.

Fill in the [brackets]. Agents can make mistakes, so review anything before it is sent, paid or deleted.

Read the original on cognition.ai ↗

Get 3 agent uses like this every morning

With the day's top AI news. Free, and one click to unsubscribe.