Well, Clojure doesn't have programmable reader macros either way.

As long as you're not dead set on specific characters you could do:

   (defun extract-from-objc (obj)
     (objc-typecase
      obj
      (%@NSDate [[[[%@NSISO8601DateFormatter @(alloc)]
                   @(init)]
                  @(stringFromDate:.) :pointer obj]
                 @(UTF8String)]s)
      (%@NSString [obj @(UTF8String)]s)
      (%@NSNumber (parse-number:parse-number
                   (objc-runtime:.:extract-nsstring
                    [obj @(stringValue)])))
      (%@NSArray (map-nsarray #'extract-from-objc obj))
      (%@NSDictionary (fw.lu:alist-string-hash-table
                       (pairlis (map-nsarray #'extract-from-objc [obj @(allKeys)])
                                (map-nsarray #'extract-from-objc [obj @(allValues)]))))
      (t (or (funcall-some (cdr (objc-pick-by-type obj *objc-extractors*))
                           obj)
             obj))))

The reason I say they're good for DSLs is exactly for re-purposing to have a special meaning. Yes, nowhere close to being as powerful as reader macros. But the nice part is everyone can and knows how to work with these primitives.

I do wish we had reader macros like CL, but I'd still totally want the built-in [] and {}.

I really like how most of the non-sexp syntax is optional in CL. Anyways, I'm also a bit ambivalent about the merits of [] vs. #(): I like how the latter syntax is more "standard", since the # dispatching macro character is used for a bunch of things. E.g. multidimensional arrays #2a() #3a() complex numbers #c(1 2) character literals #\space, pathnames #p"/foo/bar" etc.: it adds to the uniformity of the syntax, ime.

Anyways, I guess I like CL's approach where #[] #{} [] and {} are all specified as reserved to the user and then, using named-readtables, you can pick an appropriate implementation as you see fit.