How Rovo Chat embraces multi-agent orchestration

This post details the evolution of Rovo Chat's architecture from a single-agent framework to a hierarchical multi-agent system, and then to a hybrid orchestration model. It explains the rationale behind this evolution, focusing on improved subtask decomposition, domain specialization of agents (e.g., Jira Agent), the use of system tools, and different reasoning modes (Brainstorming, Tool QnA, Reasoning). The post also contrasts the hybrid model with previous single-agent and DAG-based approaches, presenting evaluation data on quality and latency improvements. Key technical contributions include the concept of domain-specialized subagents as tools, the processing of large volumes of Jira issues via iterative refinement, and the hybrid orchestrator's ability to dynamically select and execute tools one step at a time.