As long as you re-serialise that data when piping to coreutils (etc) you shouldn't have an issue.
This is what my shell (https://github.com/lmorg/murex) does. It defaults to using JSON as a serialisation format (that's how arrays, maps, etc are stored, how spaces are escaped, etc) but it can re-serialise that data when piping into executables which aren't JSON aware.
Other serialisation formats are also supported such as YAML, CSV and S-Expressions too. I had also considered adding support for ASCII records but it's not a use case I personally run into (unlike JSON, YAML, CSV, etc) so haven't written a marshaller to support it yet.
