Scanner less parsers are strictly more powerful than parsers using fixed lexers. You can literally change the language you are parsing based on what you have parsed so far. I have code running in production doing exactly that.
Generally such languages are so abstract that they lose usefulness in practice. Because something can be done, doesn't mean it should. Writing a simple recursive descent parser with a Lexer's token stream is much easier than doing bytestream ingestion on the fly, especially for a beginner.
You can even parse a definition of a parser and then continue parsing the rest of the code using that parser. Not something I have needed in production but it shows the power of non-lexing parsers.
