The AI PM Reading List for 2026: What to Read, In What Order, and What to Skip
TL;DR
There is no shortage of AI PM reading material — there's a shortage of sequenced, curated guidance on what to read first, what to read once you have foundations, and what sounds valuable but isn't worth your time. This reading list is organized by learning stage, honest about what each resource actually teaches, and explicit about what to skip.
Stage 1: Foundations (Read These First)
For understanding how LLMs actually work
Andrej Karpathy's "The Busy Person's Intro to Large Language Models" (YouTube, 60 min) is the single best technical foundation for PMs. It explains transformers, tokenization, pretraining, RLHF, and emergent capabilities at exactly the right depth — enough to reason about product decisions without pretending to be an engineer.
Follow with: Anthropic's model cards and OpenAI's system card for GPT-4 — both are readable PM-level documents about real production AI systems.
For understanding AI product strategy
Martin Casado and Matt Bornstein's essays at a16z on "The New Business of AI" (2020) and "Who Owns the Generative AI Platform" (2023) are the clearest thinking available on AI business models, moats, and where value accrues in the AI stack. They're free, they're short, and they're cited in virtually every serious AI PM strategy interview.
For understanding AI product evaluation
Eugene Yan's writing on LLM evaluation at eugeneyan.com is the best practitioner-level resource on building eval systems. His "LLM Evaluation" and "Patterns for Building LLM-based Systems" posts are required reading for any AI PM who needs to reason about quality measurement.
Stage 2: Applied Reading (After You Have Foundations)
Chip Huyen — AI Engineering (O'Reilly, 2024)
The best book-length treatment of how AI systems actually get built and shipped. PMs don't need to read every chapter — focus on the chapters on evaluation, deployment, and the sections on prompt engineering as product design.
Lenny's Newsletter AI PM issues
Lenny Rachitsky regularly interviews AI PMs and publishes structured breakdowns of how they work. The interviews are more useful than the analysis — reading 5–10 of them gives you a clear picture of what the job looks like day-to-day across different company types.
Anthropic Model Specification (public document)
Anthropic's published model spec is a masterclass in how to think about AI values, safety, and product design. It's written for a technical audience but is readable by PMs — and understanding it makes you significantly more fluent in responsible AI conversations.
The Pragmatic Engineer — AI issues
Gergely Orosz writes about AI from an engineering perspective, but his analysis of AI product decisions is among the most grounded available. His takes on what actually ships vs. what gets announced are useful calibration for PMs building product sense.
Google's PAIR Guidebook (free, online)
Google's People + AI Research team produced a free guidebook on designing AI products. It's practical, PM-oriented, and covers user trust, expectation calibration, and error handling with concrete examples. Worth a full read.
What to Skip (And Why)
Most 'AI for Managers' books published before 2024
The field moved faster than publishing timelines. Books with AI strategy frameworks written before GPT-4 are largely obsolete — the capability assumptions they're built on no longer apply. Check the publication date before investing time.
Long academic papers on alignment and AI safety (for now)
Important topics but a poor use of your first 100 hours. The relevant practical implications of alignment research can be extracted from Anthropic's public writing in a fraction of the time. Come back to the primary literature once you have product foundations.
Prompt engineering 'cookbooks' and tip collections
Knowing 50 prompting tricks doesn't make you a better AI PM. Understanding why prompts work — the model's probabilistic nature, context window dynamics, instruction following — is what builds transferable judgment. Go deep on principles, not tricks.
AI news aggregators and daily digests
Daily AI news generates anxiety and consumes time without building competency. Following two or three high-quality practitioners on LinkedIn or subscribing to one newsletter is sufficient signal. Treat AI news like market news — directional awareness, not daily consumption.
Go Beyond Reading in the AI PM Masterclass
Reading builds vocabulary. The Masterclass builds competency — with live sessions, portfolio projects, and feedback from Salesforce and Google practitioners.
Reading List Mistakes That Waste Months
Reading everything before producing anything
A reading list is preparation for doing, not a substitute for doing. After your first 20 hours of reading, stop and produce something — a PRD draft, a competitive analysis, an eval framework sketch. Reading beyond that point without production gives diminishing returns.
Treating reading breadth as learning depth
Reading 30 articles about RAG is not the same as being able to design a RAG architecture for a specific use case. Breadth gives you vocabulary; depth gives you judgment. Read two or three resources on each topic deeply rather than ten superficially.
Skipping the technical foundations in favor of strategy content
Strategy content is more readable and feels more immediately applicable. Technical foundations feel harder and less relevant. But every AI PM interview tests technical foundations — strategy fluency without technical grounding is unstable.
Not updating your reading list as the field moves
AI moves fast enough that a reading list from 18 months ago may be materially outdated. Check publication dates. Prioritize practitioner writing over academic writing for current applicability. Revisit this list quarterly.
How to Get More From What You Read
Apply every resource to a real product within 24 hours
After reading about data flywheels, spend 15 minutes thinking about how a flywheel would work for a product you use or work on. After reading about evaluation design, sketch what an eval set would look like for one AI feature. Application cements what passive reading doesn't.
Write a three-sentence summary of every resource you finish
What did it teach? What would you do differently on a product because of it? What question did it raise that you still need to answer? This forces synthesis and gives you a record of what you've actually extracted from each resource.
Share one insight per week publicly on LinkedIn
Writing a short post about something you learned forces you to clarify your thinking, positions you in the AI PM community, and occasionally attracts inbound from hiring managers and practitioners. It also creates a public record of your learning journey.
Discuss what you're reading with a cohort peer or study partner
Explaining a concept to someone else is the fastest way to discover whether you actually understand it. A weekly 30-minute conversation about what you each read that week produces more retention than three additional hours of solo reading.
Stop reading a resource once it stops teaching you new things
You don't have to finish every book. If you're 60% through a resource and the last three chapters haven't added anything, stop and move to the next one. Progress through material that challenges you — not completion of material that repeats itself.
Turn Your Reading Into Competency
The IAIPM Masterclass gives you the structured learning, applied projects, and practitioner feedback that reading alone can't provide.