10 Software and AI Books Ranked on Hype vs Value
Author: Alexandre Zajac Platform: LinkedIn
Ranking 10 software/AI books on hype vs value — one takeaway each.
The list
0. AI Engineering — Chip Huyen Your eval harness matters more than your model choice. LLM features are broken because the team never built a reliable way to measure whether an output is any good. Huyen makes evaluation the main event, which is where the hard engineering lives.
1. Designing Data-Intensive Applications — Martin Kleppmann Every database is a bundle of tradeoffs between read speed, write speed, and consistency. Read this, and you start choosing those tradeoffs deliberately, before production chooses for you.
2. Build a Large Language Model from Scratch — Sebastian Raschka Code attention by hand once, and the whole thing turns from magic into machinery you can reason about. You debug and extend models far more confidently when you’ve built the pieces yourself.
3. A Philosophy of Software Design — John Ousterhout Complexity creeps in through a hundred reasonable-looking shortcuts, each one fine on its own. Ousterhout’s fix is the deep module: a simple interface hiding a lot of capability. Aim for those and your design pain goes away.
4. Hands-On Large Language Models — Jay Alammar Embeddings do half the real work in production LLM systems: turning messy text into vectors you can search, cluster, and classify. Alammar’s visuals finally make that click.
5. The Pragmatic Programmer — Hunt & Thomas Fix the broken window. Every small mess you tolerate signals that no one’s really in charge, and the rot spreads from there. Codebases die one ignored TODO at a time.
6. Release It! — Michael Nygard Systems fail at their integration points, where they call something else and wait. Every dependency without a timeout is a thread waiting to hang you. Nygard’s stability patterns (circuit breakers, bulkheads, timeouts) are how you stay standing.
7. Understanding Deep Learning — Simon Prince The book that gets you to a real grasp of transformers and diffusion models by ordering the ideas well and cutting the math-for-math’s-sake. Approachable enough to read solo, rigorous enough to trust. The PDF is free.
8. Database Internals — Alex Petrov Your database’s read versus write performance was set by its storage engine long before you wrote a single query. Petrov takes you down to the B-trees and LSM-trees that decide it, so you can match the engine to the workload on purpose.
9. Clean Code — Robert Martin Take the good instincts (clear names, small functions) and hold the rest loosely, since plenty of the advice has aged badly. The value is understanding why engineers still argue about it, which sharpens your taste more than the rules ever will.
Notable discussion
The comments largely converged on one debate: Clean Code has aged poorly. Several readers called it dated and dogmatic, preferring A Philosophy of Software Design (deep modules, reducing cognitive load) as the more relevant pick for modern distributed and AI-heavy systems. Other books raised in the thread: Structure and Interpretation of Computer Programs, The Art of Computer Programming, Software Foundations, and Real World Haskell.
157,593 followers · 51 comments on the original post.