Playwright MCP vs CLI vs Agents: What to Use in 2026
Playwright has three ways to talk to AI: MCP, CLI, and Test Agents. Here's the decision framework an enterprise SDET uses to pick the right one for 2026.
Read the storyI'm Halmurat T. — senior SDET writing about test automation at enterprise scale. Fifteen years debugging production systems at Fortune 500 telecom, insurance, retail, and financial services. The kind of scars you can't read about in documentation.
This blog is opinionated by design. Playwright vs Selenium isn't a tie. Page Object Model is overengineered. Your flaky test isn't flaky — it's broken. If that sounds useful, stick around.
ship/ StepScout — IntelliJ plugin for Cucumber step navigation →
Subscribe to get updates straight to your inbox.

No spam · Unsubscribe anytime
Playwright has three ways to talk to AI: MCP, CLI, and Test Agents. Here's the decision framework an enterprise SDET uses to pick the right one for 2026.
Read the story →Most engineers think Claude Code has three primitives. It actually has two — skills and subagents. Here's when to use which, with token-cost benchmarks.
Playwright 1.59 ships --debug=cli and scriptable trace analysis. The first debug surface that actually works where enterprise tests fail — headless CI.
Playwright 1.59 ships a Screencast API with burned-in action annotations and chapter markers. Microsoft sells it for AI agents — the real win is CI triage.
Shared staging environments are the #1 source of flaky UI tests in enterprise teams. Here's the 5-pillar blueprint for building a controlled test environment.
How we built a Playwright API test data layer in Java — replacing UI setup that burned 150 minutes per run with 3 classes and a legacy API wrapping pattern.
Search is not available in dev mode.
Run npm run build then
npm run preview:local to test search locally.