At a glance
WHAT IT’S REALLY ABOUT
Hands-on walkthrough building production agents using Claude Managed Agents APIs
- Claude Managed Agents are production-ready API endpoints that bundle core agent infrastructure like tool execution, retries, memory, and observability so developers can compose only what they need.
- The workshop breaks the platform into four primitives—agent, environment, session, and event—then maps each to concrete API calls (list/get sessions, send events, stream events).
- A live coding demo builds a simple “deal desk” chat app that uses multi-agent coordination, MCP tools (e.g., Linear), and outcome-based iteration to produce higher-quality analyses.
- Security and control features are highlighted, including per-tool permissioning (auto vs. human approval), credential vaults that keep tokens out of the model context, and interrupts to halt unsafe behavior.
- Operational features include self-hosted sandboxes and MCP tunnels for private infrastructure, plus a developer console for live session debugging, event inspection, and memory-store editing.
IDEAS WORTH REMEMBERING
5 ideasManaged Agents replace the “agent loop” you’d otherwise have to build and operate.
Instead of implementing context management, retries, state transitions, tool harnessing, storage, and sandbox orchestration yourself, you call Managed Agents endpoints and focus on product logic and UX.
Think in four primitives to structure your architecture: agent, environment, session, events.
Agents define configuration (prompt, tools, MCP, permissions), environments define sandbox/container behavior, sessions are the running conversations, and events are the streaming interface for all interaction and lifecycle state.
Use per-tool permissions and interrupts to make agents safer in production.
You can allow low-risk tools (e.g., file read) to auto-execute while requiring explicit human approval for high-risk tools (e.g., bash execution or database/MCP access), and you can interrupt sessions to stop bad tangents.
Credential vaults and MCP tunnels enable secure enterprise integrations without exposing secrets or private services.
Vaults let you store auth tokens once and inject them at execution time without putting secrets into the model context, while MCP tunnels connect Claude to private MCP servers without exposing them publicly.
Outcomes are a practical way to enforce quality via rubric-based iteration.
By submitting an outcome (spec/rubric), the agent enters a mode where it repeatedly evaluates and revises its work against that rubric until it believes the requirements are met, improving reliability for tasks like analysis or implementations.
WORDS WORTH SAVING
5 quotesClaude Managed Agents at a high level is just a set of API endpoints that we've developed and released, um, you can go use them with any API key today, um, that give you, uh, uh, access to, uh, like scaled-ready, production-ready agent, uh, and all of the primitives around it that you can just build your own products on top of.
— Unknown
You can pick and choose whatever primitives you need, um, and ditch the rest, and then build whatever product experience you need on top of.
— Unknown
So you can decide that something like the file read tool, um, can just auto-execute, whereas something like, um, executing bash or, uh, calling your database's MCP server, um, requires explicit approval from, from your own end users.
— Unknown
And another really nice added, uh, aspect of, like, setting up all these agents is that you can also define, um, certain permission controls on a per tool basis.
— Unknown
It, it, it means that, like, Claude can help you build your own Claude.
— Unknown
High quality AI-generated summary created from speaker-labeled transcript.
