How do teams communicate the limits of an automated function?

What helps engineers and users develop a shared understanding of where a function is intended to operate and where its limitations matter? Discuss communication and evidence.