Shipping a single agent is hard. Shipping multi-agent systems that coordinate reliably is exponentially harder. When Agent A's output becomes Agent B's input becomes Agent C's decision, what happens when they disagree? How do you detect deadlock? How do you resolve conflict without cascading failures? Most agent frameworks focus on chaining: A→B→C linear pipelines. Enterprise workflows demand orchestration: multiple agents deciding simultaneously, negotiating outcomes, recovering from failures. This is where distributed systems meets agent behavior. Recent production deployments reveal new failure modes: consensus loops (agents pass back and forth endlessly), deadlock cascades (A waits for B, B waits for C, C waits for A), inconsistent views (agents decide on stale info), timeout ambiguity (fail or still thinking?). This talk explores: The multi-agent problem space (consensus, deadlock, consistency, Byzantine agents). Detection patterns (timeouts, circular dependency analysis, state divergence). Resolution strategies (voting & arbitration, process-level agreements, timeout hierarchies, rollback & replay, Byzantine tolerance). Production case studies (E-commerce order processing, autonomous research coordination, multi-agent audits). This draws from distributed systems theory but applies it directly to agentic workflows.

Most agent frameworks chain A→B→C. Enterprise workflows demand orchestration: multiple agents deciding simultaneously, negotiating outcomes, recovering from failures. New failure modes: consensus loops, deadlock cascades, inconsistent views. This talk explores detection patterns and six resolution strategies (voting, arbitration, formal consensus, timeouts, Byzantine tolerance, rollback) from distributed systems applied to agents. Case studies: E-commerce orchestration, autonomous research, multi-agent audits.