Institute of Product Management LogoINSTITUTE OF AI PRODUCT MANAGEMENT
AI Product Management

The Best AI Product Management Substacks to Subscribe to in 2026

By Institute of AI PM11 min readMay 10, 2026

The Substacks worth letting into your inbox. Each one passes a single test: a typical post leaves you smarter than you were 15 minutes ago.

Why Substack Specifically

Substack writers earn from your attention, so the good ones edit hard. The format also rewards depth: a 2,500-word post on Substack will be read; the same post on LinkedIn won't.

That said, Substack is also where think-piece inflation lives. The list below filters for writers who've shipped AI products, run AI labs, or done original research. Pundits without product experience didn't make the cut.

📬Subscribing is the easy part. The AI PM Masterclass is where you turn writing into shipped product, in 4 weekends with a Salesforce Sr. Director PM.

For Practitioners

1. One Useful Thing (Ethan Mollick)

Wharton professor running the most rigorous public experiments on what frontier models can actually do. Mollick stress-tests every new model release against real white-collar tasks and publishes the results within days.

Why AI PMs need this: When a new model drops, Mollick's post tells you what's genuinely new vs. spec-sheet noise. Required reading. Pair with our AI evaluation guide.

Subscribe to One Useful Thing

2. Latent Space (swyx & Alessio Fanelli)

The AI engineering Substack of record. Weekly news roundups, deep technical interviews, and original frameworks (the "AI Engineer" identity itself was crystallized here). Strong show notes for the podcast version.

Why AI PMs need this: If you ship with engineers, this is the room they're in. The "Rise of the AI Engineer" piece reshaped a lot of org charts in 2024–2025.

Subscribe to Latent Space

3. Interconnects (Nathan Lambert)

Nathan runs RLHF and post-training at AI2. Interconnects is the clearest writing on frontier model training: RLHF, DPO, RLAIF, scaling laws, and reasoning models. PM-readable, deeply sourced.

Why AI PMs need this: Reasoning models are the dominant product surface change of 2025–2026. Nathan tells you why they behave the way they do, and what it means for your roadmap. Pair with our RLHF guide.

Subscribe to Interconnects

For News Roundups

4. Import AI (Jack Clark)

Co-founder of Anthropic. One email a week, sober tone, focused on the research and policy stories that will become product capabilities in 6–18 months. Closing "Tech Tales" short fiction is somehow always a good metaphor.

Why AI PMs need this: The only AI newsletter that consistently respects your time. If you only subscribe to one news roundup, this is it.

Subscribe to Import AI

5. The Batch (Andrew Ng / DeepLearning.AI)

Andrew Ng's weekly newsletter. Less depth than Import AI, but broader coverage of business AI applications, education, and policy. Strong on what's happening outside Silicon Valley.

Why AI PMs need this: Useful if your product touches industries beyond pure tech — healthcare, manufacturing, education, retail. Andrew's lens is global and applied.

Subscribe to The Batch

For Skepticism and Sanity

6. AI Snake Oil (Arvind Narayanan & Sayash Kapoor)

Two Princeton researchers debunking AI overclaims with data. They cover benchmark contamination, predictive AI failures, agent demo cherry-picking, and academic paper data leakage. Not anti-AI; anti-BS.

Why AI PMs need this: Read before believing any vendor's "agent" demo or benchmark claim. Saves you from expensive bad bets.

Subscribe to AI Snake Oil

7. Marcus on AI (Gary Marcus)

The loudest skeptic on LLM scaling laws. Even if you disagree, Gary's posts surface failure cases and limitations that boosters underplay. A useful counter-balance to AI Twitter's hype loop.

Why AI PMs need this: Subscribe with skepticism toward both Gary and the people he's critiquing. Your judgment improves from holding both views.

Subscribe to Marcus on AI

For Strategy and Business

8. Lenny's Newsletter (Lenny Rachitsky)

The PM newsletter. Heavy AI coverage in 2025–2026, with deep interviews of PM leaders at OpenAI, Anthropic, Notion, Linear, Cursor, and Perplexity. Free subset is generous; paid tier unlocks the deeper benchmarks.

Why AI PMs need this: The "How AI is changing PM work" series alone is worth the price. Pair with our AI PM tool stack guide.

Subscribe to Lenny's Newsletter

9. Every (Dan Shipper et al.)

A bundled Substack with multiple writers. Dan's "Chain of Thought" series is the standout — field notes from running a real software company on top of AI tools. Practical, weird, opinionated.

Why AI PMs need this: Real workflows you can copy this week. Plus the team ships software, so the writing stays grounded in product reality.

Subscribe to Every

10. Working in AI / SemiAnalysis (Dylan Patel)

The deepest public reporting on AI infrastructure: GPU supply chains, datacenter economics, Nvidia roadmaps, model training costs, and inference unit economics. Some posts are paywalled and worth it.

Why AI PMs need this: If your product cares about cost, latency, or vendor dependency, SemiAnalysis is the only writing that goes a level deeper than the press releases. Pair with our AI cost optimization guide.

Subscribe to SemiAnalysis

For Long-Term Thinking

11. Astral Codex Ten (Scott Alexander)

Long, careful posts on AI safety, AI policy, and meta-research. ACX is where many of the more sophisticated arguments about AI risk and AI capability scaling actually get hashed out.

Why AI PMs need this: When you need to think carefully (not quickly) about a controversial topic, ACX is the model for what good slow thinking looks like.

Subscribe to Astral Codex Ten

12. Don't Worry About the Vase (Zvi Mowshowitz)

The most exhaustive weekly AI roundup on the internet. Zvi reads everything and links it. Posts are long; use the table of contents to jump to topics you care about.

Why AI PMs need this: When something happened in AI this week and you're not sure what or why, Zvi has covered it. Use as a reference, not a cover-to-cover read.

Subscribe to Don't Worry About the Vase

For Niche Expertise

13. Eugene Yan

Senior applied scientist at Amazon, writing on building production ML/LLM systems: evals, RAG patterns, system design, and team-building. The "ML system design primer" posts are reference-quality.

Why AI PMs need this: When you need to understand the mechanics of evals, retrieval, or fine-tuning at a working PM level, Eugene's the cleanest source. Pair with our RAG guide.

Visit eugeneyan.com

14. Hamel Husain

Independent ML/LLM consultant who writes the most no-nonsense posts on evals, fine-tuning, and what's actually working at his clients (which include big-name AI companies). Skip the meta; read the case studies.

Why AI PMs need this: Hamel's "evals are king" posts shaped how a generation of AI teams measure quality. Read them. Pair with our eval test case template.

Visit hamel.dev

15. The AI PM Substack (Aakash Gupta)

Aakash's "Product Growth" Substack covers AI PM specifically: interview prep, role frameworks, and weekly tactical playbooks. High frequency, free tier is generous, paid tier unlocks templates.

Why AI PMs need this: Best Substack for early- and mid-career AI PMs working on the career and craft layer. Pair with our AI PM interview guide.

Subscribe to Product Growth

Subscribing Strategy

Subscribe to four. One news roundup (Import AI), one practitioner (One Useful Thing), one technical (Latent Space or Interconnects), one strategy (Lenny's). Move email to a "Reading" folder so it doesn't crowd your inbox. Read on Tuesday and Friday mornings, in 30-minute blocks. Anything else and you'll subscribe to twenty and read none.

How to Read Substacks Without Drowning

Use Substack's Reader. The native iOS/Android Reader keeps emails out of your inbox while preserving subscriber-only access. Game-changer.

Set a "delete on Friday" rule. If you didn't read it by Friday, delete it. New posts will come. Hoarded inbox = unread inbox = abandoned subscription.

Read with a notebook open. If a post doesn't generate at least one written note, you didn't actually read it. Reading without note-taking is consumption, not learning.

Audit quarterly. Every three months, unsubscribe from anything you haven't opened twice. Your future self will thank you.

Substacks to Skip

"AI for [generic role]" newsletters with no original research. These are usually summaries of summaries. Read the source instead.

Daily newsletters from individuals. Almost no one has new insight every day. Daily output is usually rehashed tweets.

Substacks behind a paywall before a free taste. If you can't tell what you're paying for, don't pay. The good ones publish 60–80% free.

Your Subscription Plan

Today: subscribe to Import AI, One Useful Thing, Latent Space, and Lenny's Newsletter. That's it. Four sources, one focused reading window per week.

In a quarter: audit. If you've consistently read all four, add one. If you haven't, drop one. Don't add until you've earned the slot.

Reading sharpens taste. Taste turns into shipped product when you have a peer group, a structured curriculum, and accountability. Our AI Product Management Masterclass is built to be that bridge — 4 weekends, real builds, taught by a Salesforce Sr. Director PM.

Reading Sharpens Taste. Building Pays the Bills.

The best Substacks in the world won't ship your product. Our masterclass turns the best ideas in AI PM into 4 real builds in 4 weekends.