How QA Teams are Using AI to Rethink Testing

At a recent Perseus AI Roundtable, Sadia Aslam shared how her QA team has transformed their approach to software testing with Cursor. Over the past year, they’ve produced a framework that helps QA engineers build, maintain, execute, and troubleshoot automated tests more efficiently.

In this article, Sadia shares how the team approached adoption, the lessons they learned along the way, and why success with AI depended as much on their existing engineering practices as the technology itself.

Building on Existing Infrastructure

“Before onboarding Cursor, our team had already built a Playwright framework. Locators lived in one place, page actions in another, and test data was kept separate from the scripts that consumed it. We also had a catalogue of reports documenting the framework’s successes and failures.”

“When we eventually implemented Cursor, it had access to our complete framework and could make immediate connections,” Aslam explained.

Because the Playwright framework existed before the AI did, Cursor had something concrete to learn from. Rather than inventing its own patterns, it mirrored the team’s existing conventions and architecture.

Describing Behavior Instead of Writing Code

“One of the biggest day-to-day shifts for our team has been in how we interact with the testing framework. Rather than manually writing every line of code, we describe the outcome we want, and Cursor generates a Playwright test that aligns with our existing architecture.”

Because Cursor has access to the entire codebase, it doesn’t start from scratch with every request. It understands the existing page objects, naming conventions, and test patterns, allowing it to build on what already exists instead of reinventing it.

“Context will become increasingly valuable as the application grows” Aslam shared. “Engineers will spend less time expanding our codebase and more time validating behaviour.” 

Building AI Into Everyday Processes

The team’s most ambitious work goes well beyond generating test scripts; they’ve begun treating AI as a connective layer that supports the entire QA workflow.

“By establishing project rules, reusable skills, and specialized sub-agents, we’ve taught Cursor to follow the same engineering standards every time. Whether generating new tests, updating existing ones, or maintaining project structure, Cursor operates within our established guardrails rather than creating its own.

When a test fails, Cursor can analyze the error output and suggest fixes. Through integrations with tools like Jira and Playwright, it can read a ticket, generate positive and negative test cases, and execute them automatically.”

Our team has also built AI-assisted workflows that streamline reporting, update progress, and, where appropriate, prepare follow-on work when issues are identified.

Looking Ahead

Despite the productivity gains, Aslam was careful not to overstate AI’s capabilities.

“AI can be wrong, code still requires review before it’s merged, and exploratory testing remains a human responsibility.” She attributes her team’s success to their phased approach where they focus on judgment and problem solving rather than the quantity of their outputs.

Back to Insights