References: design, build, defend

Source material

The capstone introduces no new claims; it consolidates the track. These are the public sources the track’s lessons drew on, gathered in one place.

• Building Effective AI Agents, Anthropic engineering blog, December 19, 2024. The essay behind the track’s workflow-versus-agent line and its simplicity principle. If you reread one source after finishing the track, make it this one; it reads differently once you have built something.
• How we built our multi-agent research system, Anthropic engineering blog. The production account behind lesson 5’s orchestration and error-propagation material.
• Writing effective tools for agents, with agents, Anthropic engineering blog. The guidance behind lesson 4’s description and error-design standards.
• Claude Code documentation, Anthropic. Memory and configuration (lesson 2), MCP configuration (lesson 4), the Agent SDK’s subagent pages behind lesson 5, and headless mode with the CLI reference (lesson 7), in current canonical form.
• Model Context Protocol. The spec and the build-a-server tutorial that scaffold the capstone’s custom tool.
• Claude Developer Platform documentation, Anthropic. Tool use and structured output, the pages behind lesson 3.

Going deeper

• Anthropic Academy. Free, self-paced official courses that extend every surface this track touched, including courses on the Claude API, Claude Code, MCP, and subagents.

On this site

• Building with Claude (Track 22). The parts layer under this track: API calls, tool use, the agent loop, and the named agent patterns.
• Anatomy of an AI Agent Team (Track 25). The read-a-real-system track: a working multi-agent system studied role by role, including the risk gate whose criteria lesson 6 taught you to design.

Adjacent topics

• The capstone’s defense format (where judgment lives, what was traded, what breaks first, what was refused) is a general design-review shape. It applies to systems well beyond agents, and it is yours to keep.