Work public

managed-agents-research

Five progressive agent demos from single-agent to multi-agent with plan and reflect loops. Claude Agent SDK, 42 tests, per-run cost numbers.

  • agents
  • claude-agent-sdk
  • multi-agent
  • cost-analysis
  • research
Repo
github.com/mj-deving/managed-agents-research
Published
2026-05-26

What it is

A research repo that walks through the Claude Agent SDK by building five progressively harder agents. Single-agent first, then tool-using agent, then plan-and-reflect, then multi-agent coordination, then a small swarm with bounded scope. Each demo ships with tests and per-run cost numbers between 0.42 and 1.97 dollars.

The repo is a learning surface, not a product. The point is the cost-vs-capability curve you can read off the demo table.

Stack

Claude Agent SDK, TypeScript, 42 tests. Cost numbers measured per actual run.