Lots of options here, focusing on writing:

• String based API
• Text based API
• Builder-based API
  • Assume UTF8 everywhere
  • Fancy logic to determine encoding to convert to
• Perhaps import the entire say package
• Maybe define a new Display typeclass and/or helper functions to convert to Builders
• Apply similar principles to RIO.Logger, or perhaps force all output via a RIO.Logger API (probably a bad idea)

For reading: Builder is out, so it's basically guaranteed to be a Text-based API (no one wants String, right?). Big question is whether we assume UTF8 or guess character encoding (assuming the guessing).