Building Multi-Agent Systems for Finance with Strands Agents: Open-Source Patterns for Production-Ready AI Workflows

AI agents are reshaping financial services—but building reliable, composable, and auditable agentic systems remains a hard engineering problem. How do you design agents that can reason, plan, and execute multi-step financial workflows while remaining maintainable, observable, and safe?

This session presents practical, open-source approaches to building multi-agent systems for finance using:

🔓 Strands Agents — an Apache 2.0 open-source agentic AI framework for building composable agent pipelines in Python
🔓 Model Context Protocol (MCP) — an open standard for connecting agents to tools, data sources, and external APIs
🔓 AgentCore — an open runtime for orchestrating, deploying, and scaling multi-agent systems in production
We walk through two real-world financial use cases implemented entirely with open-source tooling:

Quantitative Backtesting Agent — autonomously retrieves market data, executes strategy simulations, and generates performance analytics reports
AI Fund Manager System — a supervisor-worker multi-agent pipeline for portfolio monitoring, risk assessment, and investment memo generation
Along the way, we cover the engineering fundamentals that matter in production:

Agent loop design and tool use patterns
Multi-agent orchestration: supervisor, worker, and peer-to-peer topologies
Memory management: in-context, external, and episodic memory
Guardrails and observability for financial-grade reliability
Lessons learned from real deployments
All demo code will be fully open-sourced on GitHub and available to the community after the session.