Ask HN: Do you really need a lexer/tokenizer?

Every parser/interpreter/compiler tutorial has you start with a tokenizer. Intuitively it makes sense that you would slice off this piece of work up front.

But in practice, in my hobby projects, I've found it's trivially easy to skip this step and just operate directly on characters in the recursive-descent pass. And by doing so your code is simpler, and you avoid repeating a bit of work (for example: identifying the bounds of a number literal in one pass and then fully parsing it in another pass).

Is it more necessary in full-scale compilers? Is it an old-fashioned practice that people don't really do anymore? Or am I missing something?