Note about resilient E2E testing cross teams

Notes on the team’s ideas and evaluation for introducing automated testing

• Background：A lot of time needs to be spent on manual testing, and existing tests cannot be applied immediately.
• Problems：
  • Manual testing is time-consuming and costly
  • Relies on individuals’ understanding of the product for testing
  • Automated testing usually requires writing code
  • The old UI is planned to be replaced in the future; if tests are tightly coupled to the implementation that will be a problem (but the replacement may be a long way off)
• Solutions：
  • Let AI automate tests
  • Make testers and developers jointly build specifications together
  • Describe specifications in natural language

Describe specs in natural language and have AI handle test execution. ByteDance’s open-source AI testing tool: Midscene.js🔗 seems worth trying.

• RD’s direction：

  • How to build a proper test environment? (login state, DB Seed)
  • Execute tests in JS
  • CI automated verification
  • Use local models to reduce latency and cost

• Testers’ direction：

  • Try the Chrome extension first to get a feel for it
  • Write test specs using YAML; maybe create a repo to centrally manage specs

• Experiment: as in the video, use their CLI + configure the gemini-3.1-flash-lite-preview model + cache.

• Conclusion: The fastest way to adopt it is probably to use the existing test environment directly, log into the local browser to obtain the session, then execute the test specs via Bridge Mode🔗.