Thrad.ai and AWS Bedrock automate social intelligence with multi-agent system

Thrad.ai, an advertising infrastructure platform for AI-powered LLMs, faced a signal-rich problem: their sales team spent 30-45 minutes manually researching each lead across six sources (Reddit, Hacker News, Stack Overflow, GitHub, etc.) before writing a single outreach email. To scale, they partnered with AWS to build a multi-agent system using Strands Agents and Amazon Bedrock AgentCore. The architecture features four specialist agents: a Trend Research Agent that queries six APIs for trending launches and buying-intent signals, a Search Specialist Agent that enriches prospect profiles via Wikipedia, GitHub, Lobste.rs, and Stack Overflow, an Analysis Agent that scores prospect-trend pairs (0-100) using weighted criteria, intent classification, and temporal decay, and an Email Generation Agent that drafts personalized outreach using brand knowledge and lead storage. The system uses AgentCore services (Runtime, Gateway, Memory, Observability) for orchestration and strict output validation.

The post compares two orchestration patterns—Swarm (parallel agent execution) and Graph (directed acyclic workflow)—with head-to-head benchmarks on latency, cost, and email quality. The companion repository includes a full setup guide with AWS CDK deployment, taking ~60 minutes and costing $3-5 in Bedrock model invocations. Beyond sales, these patterns apply to competitive intelligence, candidate sourcing, and market research. Governance controls (e.g., DynamoDB, Lambda, Secrets Manager) ensure production readiness. The system transforms scattered, noisy signals—like a founder asking 'What should I use for X?' on Reddit while their GitHub repo hits 2,400 stars—into a correlated, actionable prospect ready for personalized outreach.