AI-driven QA

The brief

Two brands of a global pet care company needed test automation for web and mobile. That included checking analytics, auditing CRM data and a CI/CD setup that didn’t run every suite on every change.

What I built

I built a page object model that handles regions, so US and UK variants share components and one test runs in both regions. Tests have two ways to authenticate, through the API or the browser, so they start already logged in on both apps. CRM checks against Iterable and the Central Profile API confirm where each signup came from, and custom Playwright reporters publish HTML reports to S3 and post results to Slack.

CI/CD only runs the suites affected by a change, in parallel on self-hosted runners, with full nightly and weekly regression runs. The authenticated tests went from passing 20% of the time to passing every run.

AI-driven QA

I built AI agents that work inside the development cycle:

• Test plans: agents explore the UI, map the user flows and write test plans
• Tests from work items: agents read a ticket and its acceptance criteria and write a Playwright test
• Self-healing: when a selector breaks, the agent looks at what changed in the DOM and updates the test
• Ticket flow: agents move tickets between stages based on test results
• Security and performance: agents flag OWASP top-10 issues and performance regressions in PRs
• Fixes: agents open PRs for simple fixes

This cut manual QA work a lot, sped up delivery and caught regressions before they reached main.

Looking back

The agents handle the repetitive parts of QA, like writing and updating tests.