A document summarization pipeline processes 50,000 documents per day. Each summary is independent of the others. Which architecture is most appropriate?
Answer: A leaderless swarm where workers pull from a shared queue
Document summarization is embarrassingly parallel: each document is independent, the task is homogeneous, and no inter-document coordination is needed. A leaderless swarm (or thin-orchestrator swarm) where workers pull from a shared queue delivers near-linear throughput scaling. Hub-and-spoke adds unnecessary coordination overhead. Critic-generator loops and sequential chains both increase latency without benefit for this task class.