What NOT to automate

Knowing what to leave manual is as much a professional skill as writing the code. Automating the wrong things creates a suite the team stops trusting — and a suite nobody trusts is worse than no suite.

Leave these manual

• Exploratory testing. Scripts only check what you told them to check. Discovering the unknown failure modes is human work, and it stays valuable your whole career.
• One-time verifications. Migration checks, hotfix confirmations, launch-day spot checks. The script would run once — negative ROI.
• Rapidly changing UI. If designers rebuild the screen weekly, every locator you wrote breaks weekly. Automate after it stabilizes.
• Subjective judgment. "Does this look right?", "Is this animation smooth?", "Is this error message friendly?" — assertions need exact expectations.
• Captcha and anti-bot flows. They're literally designed to defeat you. Real teams use test environments with captcha disabled.

The anti-patterns you'll be tempted by

Automating everything. A 4-hour UI suite that fails randomly gets ignored. Small and trustworthy beats big and flaky, every time.

Automating manual test cases 1:1. Manual cases include navigation and setup a machine doesn't need. Automated tests should be shorter and sharper — setup through APIs where possible, verify one thing, tear down.

Chasing 100% coverage. Coverage numbers reward quantity. Risk-based selection — the skill you already have — rewards value.