If they speak each other’s apis. R2D2 is only easier to build than C3PO if empire runs on a set of mutually intelligible standards.
It’s not voice that’s the bottleneck. Try merging 2 spreadsheets modeling the same problem by 2 independent people for example.
Yeah, seen too many "ocr of a printout of a Word document containing a photo of some printed text" situations to be sure about anything like that ☺️ human emulation mode as lcd of otherwise incompatible tech.