Claude Code: Agentic Systems At Scale

This is the capstone of the developer track. Where Level 1 made you fluent at the CLI and Level 2 made Claude Code customizable and team-shareable, Level 3 treats Claude Code as infrastructure: an agent harness to build software on rather than a tool to type into.

Over two intensive sessions you will move from a human-at-the-keyboard model to programmatic, multi-agent systems: embedding the agent loop in your own services, orchestrating workers in parallel, exposing your internal systems as tools, and running long-lived agents safely inside CI and production surfaces.

The throughline is reliability. Anyone can spawn an agent; the senior engineer's job is to know it works, keep it working, bound its cost, and contain its blast radius. You will leave with an evaluated, secured, production-grade multi-agent workflow you can defend in an architecture review.

By the end of this course, learners will be able to architect, evaluate, and operate agentic systems built on Claude Code. They will:

• Embed the Claude Code agent loop programmatically, moving beyond simple headless scripts into the SDK / agent harness inside their own applications and services
• Design multi-agent architectures: orchestrator-worker patterns, parallel fan-out / fan-in, and durable state for long-running agents
• Expose internal systems as reliable tools through MCP servers, with per-agent tool permissioning
• Build evaluation harnesses for agentic workflows: golden tasks, regression suites, and drift detection wired into CI
• Contain risk at scale: prompt-injection defense, supply-chain review of third-party skills and MCP servers, sandboxed execution, secrets handling, and token-cost governance
• Ship a production-grade multi-agent workflow with human-in-the-loop checkpoints and observability