Going Beyond Basics: Custom Agent Development Without the Starter Toolkit + Kiro
Details
### Description
Building AI agents shouldn’t mean juggling fragmented tools or patchwork prompts.
In this session, we’ll dive deep into developing custom agents from scratch—without relying on a starter toolkit—and explore how Kiro + Bedrock AgentCore revolutionizes the workflow from concept to production.
You’ll learn how to build a fully functional agent using FastAPI and Docker, meeting AgentCore’s runtime requirements while maintaining clarity and scalability.
This hands-on workshop will walk through a live example of a Financial Advisor Multi-Agent system—covering everything from dynamic prompt handling to end-to-end deployment.
**https://github.com/jayyanar/financial-advisor-multiagent**
### Key Takeaways
#### Core Learning
- Understand AgentCore’s HTTP and MCP protocol contracts
- Configure /invocations and /ping endpoints for operational readiness
#### Layered Leap: From Prototype to Production
- Natural Interface: Talk to your IDE like a co-developer
- Layered Architecture: Integrate tools and frameworks seamlessly
#### Live Demo: Financial Advisor Agent
Explore a real-world implementation featuring:
- Market Intelligence: Real-time market analysis with rolling lookback windows
- Strategy Architect: Risk-adjusted portfolio planning
- Execution Planner: Actionable, multi-horizon strategies
- Risk Assessor: Automated alignment checks and mitigations
#### From Spec to Scale
Discover how one comprehensive specification drives an entire system—built, tested, and deployed automatically through Kiro’s intelligent development flow with Hooks, GitOps, and full observability.