Agentic Testing Has a Discovery Gap Nobody Talks About
Updated April 2026.
Microsoft moved Playwright MCP into the Playwright CLI this month. Then the tutorials piled up: Hawks & Owls's Medium walkthrough wiring it into Cursor and Claude Code, alexop.dev's "Building an AI QA Engineer" with full repo, plus a Frontend Masters Agentic Playwright workshop. Every one of these is excellent. Every one answers the same question: how do I get an AI agent to author Playwright tests for me?
None of them answers the question that comes first.
Which tests should the agent be writing?
The four-stage pipeline, and the one stage nobody automated
E2E testing has four stages: discovery β generation β execution β verification . The agentic testing wave of 2025β2026 automated three of them. Generation got Playwright MCP, the Cursor sub-agent tutorials, and Mabl's Test Creation Agent. Execution got QA Wolf's managed runners and CI integrations everywhere.β¦