Introduction: Choose learning resources by purpose to avoid confusion
AI learning can be overwhelming due to the sheer amount of information. So the first recommendation is to choose resources by backward design from what you want to be able to do. For example, whether you want to master generative AI in your work, build LLM apps, or dive deeper into research will change the fastest route.
This article focuses on books, courses, and communities, highlighting resources that translate well to real work. There will be technical terms, but we will explain them as plainly as possible.
First decide: Your learning goals (3 types)
1) "User" — Mastering Generative AI in the workplace
Prompt design, information organization, writing and document creation, research, and considerations for internal adoption. This is an area where non-engineers can achieve results easily.
2) "Builder" — Creating LLM apps / automating tasks
APIs, RAG (internal document search + generation), evaluation, operations, security, etc. This is a field with high growth potential for engineers and PMs.
3) "Understanding" — Understanding the mechanics to make informed judgments
Transformers, training and inference, RLHF, data, biases, safety, etc. This improves decision-making and design reviews.
Recommended books (by use case)
For those who want to deliver results first in business/real-world work
- Practical books in the 'Textbook for Implementing Generative AI' series: These cover how to roll out internally, how to create rules, and common failure cases. (Since publishers release multiple titles, look for ones whose table of contents include "Internal Guidelines," "Use Cases," and "Risk Management".)
- Books on Writing × AI: Before prompts, good questions, good summaries, and good structure matter. Books that teach writing and editing frameworks form the foundation for leveraging Generative AI.
Engineers: Directly tied to LLM app development
- Designing Data-Intensive Applications (DDIA): Not strictly an AI book, but it covers design fundamentals (reliability, scalability, data design) you must master when putting RAG or agents into production.
- The ML classics (e.g., Hands-On Machine Learning): Even with the prominence of LLMs, fundamentals like features, evaluation, overfitting, and data leakage remain common. Concepts of classification, regression, and validation help with designing LLM evaluation.
- LLM development books that don’t rely solely on prompts: Cover RAG, function calling, evaluation, guardrails (safety), and observability.
