Multi-agent social intelligence with Strands Agents and Amazon Bedrock

This article demonstrates how Thrad.ai deployed a multi-agent system with Strands Agents and Amazon Bedrock AgentCore to automate prospect discovery and personalized email generation by correlating signals across multiple social sources.

• Four specialized agents handle trend research, prospect enrichment, scoring, and email generation with strict output validation
• Swarm orchestration enables dynamic agent handoffs with shared context; Graph orchestration uses fixed workflows with parallel entry points
• Graph pattern achieves 32s latency and $0.06 cost per prospect versus Swarm's 45s and $0.08, while Swarm produces higher-quality emails (8.2 vs 7.6)
• Scoring uses weighted criteria (topical alignment, timing, engagement, intent, data quality) with temporal decay and ICP matching
• Amazon Bedrock AgentCore provides Runtime, Gateway, Memory, and Observability services for production deployment with session isolation and distributed tracing
• Governance controls include policy gates via conditional edges, scoped tool access per agent, and Swarm safety bounds to prevent runaway behavior

The patterns extend to competitive intelligence, candidate sourcing, and content curation, with a companion repository available for deployment and customization.