https://blog.anytest.dev/enAn engineering journal about QA when AI agents explore your app and write the tests. Concrete techniques, named trade-offs, primary sources.https://blog.anytest.dev/posts/anytest-qa-team-capacity-roi/https://blog.anytest.dev/posts/anytest-qa-team-capacity-roi/Wed, 07 Oct 2026 00:00:00 GMTA measured adoption plan for QA engineers and their bosses, with a benchmark protocol and transparent ROI simulations for teams of one, two, three and five.https://blog.anytest.dev/posts/retry-budget-not-a-repair/https://blog.anytest.dev/posts/retry-budget-not-a-repair/Wed, 07 Oct 2026 00:00:00 GMTA test that passes on its second attempt still failed once. Keep that fact visible, price the repeated work, and give every flaky test an owner.https://blog.anytest.dev/posts/failure-evidence-before-rerun/https://blog.anytest.dev/posts/failure-evidence-before-rerun/Wed, 07 Oct 2026 00:00:00 GMTThe final screenshot rarely tells you how a test arrived there. Preserve the sequence, network context and assertion before another run changes the scene.https://blog.anytest.dev/posts/separate-context-is-not-separate-account/https://blog.anytest.dev/posts/separate-context-is-not-separate-account/Wed, 07 Oct 2026 00:00:00 GMTBrowser contexts isolate cookies. They do not stop two tests changing the same server record. Give parallel writers distinct accounts and data.https://blog.anytest.dev/posts/green-checkmark-not-proof/https://blog.anytest.dev/posts/green-checkmark-not-proof/Wed, 07 Oct 2026 00:00:00 GMTAI agents can now write your end-to-end tests. They inherit the oldest bug in testing: asserting on what is easy to see instead of what must be true. The fix is an independent oracle.https://blog.anytest.dev/posts/locator-is-a-contract/https://blog.anytest.dev/posts/locator-is-a-contract/Wed, 07 Oct 2026 00:00:00 GMTGenerated tests love CSS paths and XPath. Those selectors describe where a button sat on the day the code was written, not what the button is. Here is the vocabulary that survives a redesign.https://blog.anytest.dev/posts/agent-browser-needs-a-fence/https://blog.anytest.dev/posts/agent-browser-needs-a-fence/Wed, 07 Oct 2026 00:00:00 GMTA script cannot be talked into anything. An agent can. When your QA automation reads web pages, every page becomes a potential instruction source, and OWASP has a name for the attack.https://blog.anytest.dev/posts/what-agentic-qa-actually-saves/https://blog.anytest.dev/posts/what-agentic-qa-actually-saves/Wed, 07 Oct 2026 00:00:00 GMTThe honest ledger has three lines: authoring time you no longer spend, review time you now must spend, and repair time for the tests that rot. The savings are real when the first line beats the other two.