Currently, many test cases at GoTo are manual. Although automated tests exist, they require Software Development Engineers in Test (SDETs) to write and execute these tests on the cloud. This round-trip process takes a significant amount of time, especially during Functional Certification Testing (FCT), where it can take a team one to two days to complete the entire test suite.

This testing approach isn't scalable. For instance, if the GoTo app expands into new countries, localization efforts will double the size of the test suites. Additionally, many tests rely heavily on specific flows. If a new screen or dialogue unexpectedly appears due to a remote configuration or advertisements, tests often fail because they encounter scenarios not previously covered.

To achieve this, I leveraged Transformer-based Large Language Models to build intelligent UI Automation Agents. By embedding these AI-driven agents early in our software development cycle, I automated complex UI testing scenarios, reducing reliance on manual test cases and significantly shortening the feedback loop. This approach enabled us to quickly detect UI regressions, even in dynamic scenarios involving remote configurations or unexpected screens, ensuring greater stability and reliability in the main branch and final product.