What keeps Apache Kafka from eating the world?

One Size Fits All solutions usually don't actually. An architecture with Kafka has overhead, like any architecture. Sometimes that overhead is more costly than the supposed benefits.

So the reason it doesn't eat the world in spite of best effort by resume-padding "architects" is that some of us successfully push back against using it to shuffle a few GB of data per day around, where a vanilla Postgres, MySQL or often even sqlite db will do.

It doesn’t have a mouth.