How much of testing can you actually automate without losing the human eye?

I have been thinking about where automation genuinely helps versus where it just gives you false confidence. On the backend side, checking that an endpoint returns the right data for a given input is repetitive and mechanical, exactly the kind of thing a machine should own so people can focus on the judgment calls.

What I keep running into is teams that automate everything and then stop actually looking at their product. The tests are green, so nobody notices the flow that technically works but feels broken to a real person. Automation catches the regressions it was told to watch for. It does not catch the stuff you never thought to check.

I have been comparing different api testing tools lately to handle the mechanical layer, and the honest conclusion is that the tool is only half of it. The other half is a human deciding what is worth testing in the first place, and then still using the thing like a real user would.

Curious how others here balance it. Do you lean heavily on automation and trust the green checkmark, or do you keep a manual pass in the loop no matter what?