In a language with lenses, you can delimit where your models are mutated. In a language that lets you treat composable effects in a generic way, you can e.g. split up your variables explicitly with ReaderWriterState, making it clearer which serves which purpose.

The kind of things you might be tempted to use "magic" or global variables for in a lesser language - the "cross-cutting concerns" - you can instead handle in normal code as a generic kind of context. E.g. an audit trail can just be Writer. Or actions that need access to a database session can just be another kind of context. I'm working on such a thing right now (see tpolecat's doobie for a public example of the same kind of thing) - whether a function accesses the database or not is visible right in its type, but it's easy to compose functions that do. The types ripple up, and in the end I'm probably going to do session-creation-in-view - but in a principled way, using the ordinary type system, visible in the code - with Spray, my web route definitions are just another ordinary bit of Scala code (whereas in most languages I would define routes in an external config file, or at best a config object a la Django - adding complexity).

Heck, error handling is a perfectly good example of this. In a less expressive language, the only way around the boilerplate of handling errors where they occur is unchecked exceptions (or effectively-unchecked exceptions, as in "throws Exception"). In a language with good generic context support, the possibility of failure is just another kind of context that we compose with the same generic tools as any other, and the control flow becomes something sensible again.