Descriptions:
An Anthropic engineer leads a live coding workshop at a launch event, walking developers through building a production-ready multi-agent application using the Claude Managed Agents API — Anthropic’s suite of hosted primitives for scalable agent deployment. The session covers the three core building blocks: configuring agent templates with tailored system prompts and tool access, defining execution environments with customizable sandbox behavior (including newly announced support for self-hosted infrastructure via Cloudflare, Modal, and Vercel), and connecting MCP servers with per-user credential injection.
The hands-on portion centers on building a multi-agent M&A analysis tool called “Deal,” which delegates work to specialized subagents — one focused on macro industry trends, another on financial analysis — each with its own tool access and MCP connections. The presenter works through real implementation errors live, illustrating how the sessions API, agent IDs, and environment IDs fit together to launch and manage agent conversations.
The session also tours Anthropic’s developer console observability views, which provide live debugging of agent activity in production. Developers leave with a working mental model of Claude Managed Agents and code they could realistically deploy the same day.
📺 Source: Claude · Published May 21, 2026
🏷️ Format: Hands On Build
