Comment by ljm

Fundamentally it boils down to knowing the person you're talking to and how they deal with feedback or something like rejection (like having a PR closed and not understanding why).

An AI agent right now isn't really going to react to feedback in a visceral way and for the most part will revert to people pleasing. If you're unlucky the provider added some supervision that blocks your account if you're straight up abusive, but that's not the agent's own doing, it's that the provider gave it a bodyguard.

One human might respond better to a non-violent form of communication, and another might prefer you to give it to them straight because, like you, they think non-violent communication is bullshit or indirect. You have to be aware of the psychology of the person you're talking to if you want to communicate effectively.