Automatic tests are not the only way to test software, nor are they sufficient. Whether some software can actually be used is not something that's completely formalizable and so you can't test it automatically. You always need a human in the loop in the end, and alert is a legitimate tool to effortlessly display some additional information on internal state during an interactive user test.

> ad hoc in-browser debugging

Ah, if only there was a single word - a verb perhaps - that developers could use to colloquially refer to this process!

Not testing the code, fidling with the stuff they are learning.