Most teams begin their AI journey with a single assistant: generate a test case, explain a failure or create an automation script. But what happens when one assistant becomes a team of specialized agents?

In this session, I will demonstrate an agentic QA workflow in which multiple AI agents collaborate across the testing lifecycle. A Requirements Agent identifies risks, a Test Data Agent prepares the environment, a UI Agent creates Playwright tests, an Execution Agent investigates failures, and an independent Reviewer Agent evaluates the quality of the results.

At first, the workflow appears highly productive. It generates tests, executes them and reports results with minimal human involvement. But the demonstration reveals a different reality: agents repeat work, propagate incorrect assumptions, approve each other’s mistakes, consume unnecessary tokens and sometimes optimize for a green pipeline instead of meaningful product quality.

We will examine the role boundaries, handover contracts, permission models, quality gates, observability and human approval points required to make a multi-agent QA workflow reliable.