Your Sub-Agents Are a Distributed System. Treat Them Like One. - Girish Konda, Microsoft

Why watch

This advanced talk explains how AI agent orchestrators create hidden distributed systems with classic failure modes: undetected partial work, hallucinated facts propagated as truth, invisible rate limits, and duplicate side effects from retries. Girish Konda maps each failure to distributed-systems theory and presents his SAFE hand-offs framework from MCP Pacemaker. Developers building production AI agents with multiple sub-agents will learn architectural patterns for reliable, consistent agent coordination.