How Claude Code Built My First Claude Managed Agent (From Scratch)

Source: Joe Njenga, How Claude Code Built My First Claude Managed Agent (From Scratch) (Medium, 2026-04-09)

Key takeaways

  • Concrete build log — Njenga uses Claude Code itself to build a production-ready Managed Agent on Anthropic’s platform.
  • Published one day after Anthropic’s Managed Agents public beta launch.
  • Demonstrates the “days, not months” time-to-production claim with a real example.
  • Good reference for tech leads evaluating Managed Agents vs existing orchestration.

The setup

On 8 April 2026 Anthropic released Managed Agents as a public beta. Njenga sat down with Claude Code and built a production agent using the new service. Article is the walkthrough.

What Njenga built

A production-ready Claude Managed Agent covering:

  • Agent definition (tasks, tools, guardrails)
  • Deployment to Anthropic’s infrastructure
  • End-to-end testing
  • Integration into an existing workflow

What the walkthrough validates

  • The “prototype to production in days” claim holds for simple agent shapes
  • Claude Code itself is a capable tool for building Managed Agents (meta-loop)
  • The API surface is small enough that one engineer can internalise it in a session

What it doesn’t address

  • Multi-agent coordination (that’s behind a request form at claude.com/form/claude-managed-agents)
  • Advanced governance (delegation-policy visibility, per-model attribution — see the Regulated AI page)
  • Integration with existing non-Anthropic orchestration (LangGraph, CrewAI, OMC) — Managed Agents is a replacement, not a layer on top

The broader comparison to make

When reading this article, hold it alongside:

If you’re currently building on third-party orchestration, the decision to stay or migrate is the key question this article helps you answer.

Built by Force Information Systems · Harris Computer · Constellation Software. Licensed under MIT.