References
Primary sources come first. Each one states which lessons use it. Treat secondary sources as supporting reading only.
Primary sources
| Source | Used in | What to take from it |
|---|---|---|
| Building effective AI agents, Anthropic | 0001, 0002, 0003, 0007 | Workflows versus agents, the five workflow patterns, and the tool appendix on the agent-computer interface. |
| Effective context engineering for AI agents, Anthropic | 0004, 0008 | Context rot, the attention budget, just-in-time retrieval, compaction, note-taking, and sub-agent summaries. |
| Using function tools with human in the loop approvals, Microsoft Learn | 0005 | Approval pauses a run, the caller resumes it, approvals must bind to a pending request, and credential scope in production. Confirm current release status. |
| Demystifying evals for AI agents, Anthropic | 0006 | Graders (code, model, human), capability and regression evals, pass@k and pass^k. Verify metric definitions against the original. |
| Handling stop reasons, Claude Platform docs | 0001 | The tool_use and end_turn stop reasons and how the loop terminates. |
Secondary and reference
| Source | Caveat |
|---|---|
| Writing effective tools for agents, Anthropic Engineering | Read in full before relying on specifics. Recommended for lesson 0003 extensions. |
| Microsoft Agent Framework overview | Vocabulary reference for agents versus workflows when comparing frameworks. |
| Biweekly Engineering summary of Building effective agents | Third-party summary. Use only to reinforce the primary source. |
| When not to build AI agents, mer.vin | Not verified against Anthropic's pages. Treat its decision flow as a hypothesis. |
Open items
Observability for LLM calls is not yet covered. The OpenTelemetry GenAI conventions are still in development status, so pin a version before teaching them. Multi-agent coordination beyond the sub-agent pattern also needs a primary source.