namespaces-reference: ns vs in-ns

Reading https://clojure.org/reference/namespaces I was a bit surprised how it describes setting up a namespace when REPL-ing (first section, third paragraph):

The best way to set up a new namespace at the top of a Clojure source file is to use the ns macro.
...
At the Repl it's best to use in-ns, in which case the new namespace will contain mappings only for the classnames in java.lang. In order to access the names from the clojure.core namespace you must execute (clojure.core/refer 'clojure.core).

What’s the rationale behind the it's best to use in-ns?
However nice to read what ns does behind the scenes to turn a ‘bare’ namespace into something ‘useable’, it’s not something that seems to be better than ns ?

Start with the namespaces reference at https://clojure.org/reference/namespaces, especially the first section and third paragraph. Compare the documented uses of ns and in-ns, then clarify the rationale for recommending in-ns at the REPL; done means the wording answers why it is preferred.