What 15 years in QA taught me about shipping
The bugs that hurt are never the ones in the test plan. Some patterns I keep seeing across teams, products, and decades.
I have been testing software since before "shift left" was a phrase people said out loud in meetings. A few things have held up.
Coverage is a proxy, not a goal
Ninety percent line coverage on a payment flow with no negative-path tests is worse than forty percent with them, because the number buys false confidence from people who do not read the report.
The test suite is a product
It has users — the developers waiting on it. If it is slow, flaky, or unclear about why it failed, they will route around it. Every flaky test you tolerate spends credibility you will need later.
Automate the boring, judge the rest
Automation is very good at telling you something changed. It is bad at telling you something is wrong. That second part is still the job.