Could agentic AI democratize testing so that non-testers can spin up quality checks via natural language?