keyuchen21/agentic-engineering-handbook

What it solves

Building production-grade AI systems now requires mastering agents, tool use, MCP, memory, long-running workflows, coding agents, agent harnesses, evals, and safety. That knowledge is scattered across OpenAI blogs, Anthropic engineering posts, SDK docs, cookbooks, and research papers. This handbook consolidates 218 curated resources into one structured learning roadmap, with the goal of helping you become a world-class Agentic Engineer.

How it works

The handbook is organized as a phased learning roadmap. Phase 0 walks through the agent loop — the core idea of model + tools + one loop — with five runnable Python examples (v0 through v4) that build up from a bash-only agent to one with subagents and skills. Phase 1 covers agent foundations: mental models like leverage, a four-question checklist for deciding between a workflow and an agent, and prompting guidance from OpenAI, Anthropic, and Google. Phase 2 dives into the MCP and tool ecosystem. Phase 3 focuses on context, memory, and skills. Phase 4 covers harnesses and long-running agents, and later phases extend into evals, production systems, and advanced coursework such as agent training and search.

Each phase lists Read First and Then Read resources from vendors like OpenAI, Anthropic, Google, Cloudflare, and Hugging Face, followed by a Build Exercise that turns reading into hands-on practice. There are also Applied Practice Tracks for coding agents, security, code review, or SRE, and a Full Reading Table with P0/P1/P2 tiers for quick reference.

Who it's for

The handbook offers different starting points for four audiences: newcomers to agents (start at Phase 0 and follow the core phases), developers already building LLM apps (start at Phase 2 or 3 and fill gaps), people trying to build projects (use the Build Exercises and practice tracks), and anyone looking for references (jump to the Full Reading Table).

Highlights

• 218 curated resources consolidated into one structured roadmap
• A phase-based learning path (Phases 0-7) covering the agent loop, MCP, context/memory/skills, harnesses, evals, and production systems
• Runnable Python code in Phase 0's agent-loop lab: five versions showing how an agent grows from bash-only to skills-enabled
• Concrete Build Exercise prompts and Applied Practice Tracks for coding agents, security, code review, and SRE
• A Full Reading Table with P0/P1/P2 reference tiers