For the AI engineer — grounded in the canon
Build AI systems that actually ship — read from the field's best books, not the hype cycle.
The AI-engineering canon is exploding and half of it is marketing. This is the reconciled version: what the people who build reliable, evaluated, production AI actually agree on — foundation-model apps, MLOps, evaluation, LLMs — cited per claim, honest about where the field disagrees, light enough to carry. A bicycle for your craft. You still pedal.
How a guide actually reads
MisconceptionA better model is what separates good output from bad output.
RealityPrompt structure, explicit format specification, and few-shot examples often close the gap without touching the model — exhaust prompting first (ai_engineering, data_analysis_with_llms).
from the “Prompt Engineering Quality” section of Build AI Systems That Ship — grounded in AI Engineering: Building Applications with Foundation Models, Data Analysis with LLMs, Natural Language Processing with Transformers Building Language Applications with Hugging Face, AI Agents and Applications (with LangChain, LangGraph, and MCP)
Why this is different
Grounded
A free summary is one voice with no receipts. Here, every claim traces to the books behind it — we read the canon so you don’t have to, and show the sources.
Aware
We take the comprehensive view — naming where the field broadly agrees and where the outliers are. On the outliers we take a position, weighed by the type and quality of the evidence behind it: the author’s, or our own research.
Essential
Not exhaustive, not thin — only what’s load-bearing, ordered as a journey from foundations to the summit. Light enough to carry.