A Four-Layer Multi-Agent Architecture for Automated Journalism: Event-Driven Orchestration with Hybrid Context Management
Image credit:Abstract
We present a four-layer architecture for automated journal- ism that addresses unbounded context growth, agent coordination com- plexity, and quality assurance at scale. The system comprises Layer 1 (Observability), Layer 2 (Specialized Agents), Layer 3 (Event-Driven Or- chestration with Listener-Aware Communication), and Layer 4 (Hybrid Context Management with RAG). The architecture is implemented using Deep Agents, a LangGraph-based harness employing asynchronous mes- sage queues, agent pooling, domain-specific skills loaded via progressive disclosure, and remote MCP servers for external tool integration. The hybrid context strategy combines semantic compression (avg 56% reduc- tion) with RAG-based retrieval for unbounded source handling, while subagent spawning isolates context across delegated tasks. Evaluation on 500 stories sampled from a 10,000-article corpus across 5 categories demonstrates a 97.4% pipeline completion rate, average processing time of 135 seconds per story, and an average fact-check score of 0.93. Baseline comparisons against a single-agent RAG pipeline show a 25.4% improve- ment in fact-check scores.
