Appearance
00.5 — Recommended Learning Sequence
This is a suggested pacing, not a rigid schedule. Adjust to your actual starting point (see 02-competency-map.md for how to assess that).
Sequence for someone starting from a strong AI-engineering base (most likely case)
| Stage | Parts | Focus | Rough pace |
|---|---|---|---|
| 1 | Part 1 | Skim for gaps; go deep only on FastAPI, PostgreSQL, Redis, and Testing if rusty | 1 week |
| 2 | Part 2 | Read fully once; this grounds "why" for everything after | 1 week |
| 3 | Part 3 | Full depth — this is the conceptual core the rest of the book builds on | 2 weeks |
| 4 | Parts 4–6 | Full depth — architecture/internals of LangChain/LangGraph/LangSmith, build the progressive examples | 3 weeks |
| 5 | Part 7 | Full depth — this is where "it works on my machine" becomes production | 2 weeks |
| 6 | Parts 8–9 | Full depth — reliability and security, non-negotiable for enterprise deployment | 2 weeks |
| 7 | Parts 10–11 | Full depth — enterprise integration and system design, build at least 2 of the Part 11 designs on paper | 2 weeks |
| 8 | Parts 12–15 | Full depth — the customer-facing layer that differentiates FDE from AI engineer | 2 weeks |
| 9 | Part 16 | Build at least 3 projects fully, including the full-engagement one | 3–4 weeks |
| 10 | Part 17 | Interview prep, mock system design sessions | 1–2 weeks |
Total: roughly 4–5 months at a sustained, serious pace alongside other work. Compressible if you already have strong Parts 1, 4–6.
Sequence for someone with a general software background but new to AI engineering
Go strictly front-to-back, Part 1 through Part 17, without compressing Parts 1–2. The AI-specific material in Parts 3–6 depends on genuinely understanding embeddings, transformers, and LLM behavior first, not just the API surface.
A rule for every stage, regardless of background
Don't move to the next part until you've done that part's exercises and can answer its FDE scenario questions out loud, to an imagined skeptical customer, without notes. This book's chapters are structured so each one ends with exactly this check. If you can't answer them, that's signal to reread the "why" and "internals" sections, not to push forward and hope it clicks later.
Milestones worth pausing at
- After Part 3: you should be able to explain, from memory, why RAG exists and when it's the wrong tool.
- After Part 6: you should be able to trace a multi-agent LangGraph run end-to-end in a LangSmith trace and explain every hop.
- After Part 9: you should be able to look at an agent architecture and immediately spot at least 3 concrete injection/excessive-agency risks in it.
- After Part 11: you should be able to whiteboard an enterprise RAG system from a one-sentence customer prompt, unaided.
- After Part 16: you should have at least one project you could demo live to a real prospective employer or client.
Next: begin Part 1 — handbook/01-software-engineering/01-python-for-ai-engineers.md.