I disagree unless the tests are reasonably high level.
Half the time the thing you’re testing is so poorly defined that the only way to tighten that definition is to iterate.
In this sense, you’re wasting time writing tests until you’ve iterated enough to have something worth testing.
At that point, a couple of regression tests offer the biggest bang for buck so you can sanity check things are still working when you move on to another function and forget all about this one