Study notes · 7 paths · about 20 weeks in all

Learning paths

Courses I put together for building agents that hold up in production. Each one sequences the best free and paid material I could find, with labs, checkpoints and a capstone. They run roughly in this order: protocol first, then building, then operating.

1. 01MCP Learning PathModel Context Protocol from concepts to shipping and contributing to an MCP server.6 weeks
2. 02MCP Apps and SkillsThe newest MCP extensions: interactive UIs served by MCP servers, and Agent Skills served over MCP.1 week
3. 03Context EngineeringDecide what goes into the model's window and when: skills, sub-agents, compaction, memory and just-in-time retrieval.3 weeks
4. 04Agent-to-Agent (A2A)MCP connects an agent to its tools; A2A connects agents to each other. Learn the protocol and when to use which.2 weeks
5. 05Agent EvaluationTest whether an agent picked the right tools, took a sensible path and finished the job, not just whether the last answer read well.3 weeks
6. 06Agent ObservabilityTrace model calls, tool calls, tokens and cost across a multi-agent system with OpenTelemetry's GenAI conventions.2 weeks
7. 07Agent SecurityPrompt injection through tool output, least-privilege tools, agent identity, and red-teaming agents before they ship.3 weeks