Playwright 1.64: a completed WebMCP call can still return an error. These tools are website actions for AI. Experimental support. Check returned errors and failed calls. Guide: anton.qa/blog/posts/test-pag… #Playwright

Playwright 1.64 adds experimental WebMCP support. Does your site offer actions to AI? Playwright can list and call them. See what changes: anton.qa/blog/posts/test-pag… #Playwright

Try both report.pdf and report.PDF when testing file names. The tutorial's sample accepts one and rejects the other. Five small checks make the missed case visible. Full example: anton.qa/blog/posts/first-ag… #AI #Testing

First agent testing task: ask Claude Code to test a PDF filename helper. report.PDF should pass, but the sample code rejects it. Ask for the failed test and exact values. Stop before the fix. Full example: anton.qa/blog/posts/first-ag… #AITesting

Choosing a Playwright page-object pattern? A field can store a fixed search. A method can accept a changing product name. See the small comparison: anton.qa/blog/posts/playwrig… #TestAutomation

A stored Playwright locator can find a button after the page redraws. It describes how to find the element. A getter is not required for that. Compare both approaches: anton.qa/blog/posts/playwrig… #Playwright

Reduced motion asks for less movement. Does the page respond? This practice page should change immediately. Its broken version still takes two seconds. Try both: anton.qa/blog/posts/test-dar… #Playwright

Dark mode is on. The page stays white. This Playwright example checks the page's colours. Remove the dark styling rule and watch the check fail. Run it: anton.qa/blog/posts/test-dar… #Playwright

Let's save the clues from a failed Playwright test. Playwright runs browser tests. A retry is another attempt. Compare two recording settings and inspect the original failure. Run the example: anton.qa/blog/posts/save-fai… #Playwright

Microsoft ran 507 business jobs through AI agents, 20 times each. Worked once: 65%. Worked all 20 times: 25%. Their paper: "One Success Isn't Reliability." Testers already had a word for that gap. The ten-run check: anton.qa/blog/posts/does-you…

Your regression suite is a museum. Old tests nobody visits, guards nobody questions, a bill everyone pays. 5 questions decide delete vs keep. Last audit retired a third of a suite. Nothing was missed. Code walkthrough: anton.qa/blog/posts/regressi…

Shift-left made developers own testing. Most have never opened Playwright. They still have to ship tested code. The 5 questions every developer asks when handed testing, answered with code: anton.qa/blog/posts/handed-t…

AI can write a clean test that proves nothing. Before keeping one, check its risk, result, data, failure message, and release value. The full seven-check review card: anton.qa/blog/posts/review-a…

Your build is green. One of your tests failed five minutes ago. It passed on a retry, ran alone, and the race condition shipped. Playwright even labels it: flaky, not passed. Nobody reads that word. anton.qa/blog/posts/test-ret…

Your login helper exists twice. One waits for the dashboard. One waits for the error. The rest is copy and paste. The fix Playwright already uses: login(user, pass, { waitFor: 'error' }) anton.qa/blog/posts/reuse-pa…

Should page objects assert? The rule that survived a good debate: - business checks → in the test, value written out - technical guards → in the page object, as waits Code + migration path: anton.qa/blog/posts/where-te…

Your tests work. Their structure is stuck in 2015. Five signs: shared parent classes, manual waits, hidden locators, repeated setup, and hidden checks. The modern Playwright pattern replaces them without a full rewrite. anton.qa/blog/posts/modern-p…

An AI model vanished for 19 days this summer. By government order. Pinned teams fixed it in one line. Twice. The 3-rule system — pin, spare, re-verify — with the code: anton.qa/blog/posts/your-ai-… #AITesting #QA