UI testing for Power Platform has always been painful - lack of tooling, a lot of upfront effort, constant maintenance and tests that break with every form change. This session shows a different approach: using LLMs to generate stable, maintainable test code from natural language scenarios.

Why UI-level tests over unit tests? When AI writes code, it writes tests that pass - adjusting both together until you get green checkmarks without real validation. User behavior simulation is the honest feedback loop: tests describe what users expect, not what the code does.

The workflow: an LLM with Playwright MCP explores your app, identifies testable scenarios, and generates Gherkin test plans. A GitHub Copilot agent converts those into Playwright code that runs without AI in your pipeline - deterministic, no inference costs. When tests fail, agent mode analyzes failures and proposes fixes.

Gherkin scenarios are implementation-agnostic - they describe behavior, not selectors. This makes them stable when UI changes, and lets them serve dual purposes: executable specifications that guide AI coding agents, and test definitions that validate the result. For new apps, you generate scenarios from specifications before any code exists - test-driven development with AI-generated tests as quality gates.