We’ve spent (and paid) a lot fighting flaky end-to-end tests: selectors break whenever the UI changes, suites become brittle, and pipelines go red for the wrong reasons.

Bugninja is our attempt to fix that. It’s an open-source CLI on top of Playwright where you write tests in plain English, and an AI-powered agent handles navigation, element finding, and context. Runs are recorded and can be replayed, with optional self-healing when things change. pip install bugninja

# Define a test [task] name = "login_test" description = "Navigate to login page and log in with provided credentials"

bugninja run --task login_test bugninja replay --task login_test --healing We open-sourced it this week and would love feedback: Does this approach actually solve E2E pain, or just move it around? Would you trust AI in your CI/CD pipeline? For big suites, what’s the blocker we should tackle first? Repo: github.com/bugninja-ai/bugninja

Unfortunately not useful for the projects I am writing tests for in my everyday. Feeding data into bigwig API's is a no-no for many sensitive applications.

Also, how do you handle running out of context? I've found that testing with AI is fraught with issues where the model or agent loses its understanding of constraints and begins hallucinating.