When should we NOT automate a test in our regression suite?

"100% automated regression" is a productivity trap. Here's my "don't automate" list, with the reasoning you can hand to leadership: **1. Anything that changes weekly.** Onboarding wizards, marketing pages, promotional banners, A/B tested flows. Automation cost > manual cost when the target moves faster than your suite can adapt. **2. Visual polish and UX judgment.** "Does this feel right?" is not automatable. Snapshot tests catch pixel diffs, not "this animation is jarring". Reserve human eyes for perceived quality. **3. Exploratory testing of new features.** Automation confirms known behavior. Exploration discovers unknown behavior. Automating exploration = confirming what you already thought, missing what you didn't. Every new feature deserves 2–4 hours of unscripted use before you write a single automated test. **4. Complex setup / rare edge cases.** A bug that requires "database in state X, user 3 days into trial, timezone Pacific, feature flag Y half-rolled" — 4 hours to automate, runs once a year. Manual repro + monitoring is cheaper. **5. Physical / sensory / hardware.** Camera quality, audio pipelines, haptics, print output, physical devices. Automation exists but has diminishing returns; humans catch what synthetic checks miss. **6. Security testing (mostly).** Automate the known-CVE and dependency scanning. Manual pentesting for logic flaws — the interesting bugs don't show up on Burp active scan. **7. Anything that would take longer to automate than to run 100 times manually.** Simple math: hours-to-automate × maintenance-multiplier vs (manual-time × frequency × horizon). If the automation ROI is negative in 12 months, don't build it. **8. Regulatory / compliance sign-off flows.** Auditors want a human signature on the QA record. Automation supplements but doesn't replace. **Reframe the leadership ask:** "100% automated regression" is the wrong metric. The right metrics are: - **Time to production-ready feedback** (goal: <30 min from PR to green build). - **Escaped defect rate** (goal: <1 P1 per quarter reaches production). - **Test suite maintenance cost** (goal: <20% of QA capacity). Meet those and nobody cares if you're at 60% or 90% automation. Chase 100% and you'll ship 30% of the tests and hate life.