I mean, it can be an API using a format easily put into human speech, and then machine-recognized. Said format would be a language, or even a code, intended for human-machine interaction via speech, like there are codes intended for error correction in various media with varying nature of errors.
So that humans would be able to give voice commands almost in natural language.
Only this wouldn’t be such groundshaking news, older Internet protocols like SMTP and FTP already are human-readable.
This also wouldn’t cost nearly as much as the title implies.