Every "best AI blogs of 2026" list opens with the same five entries: OpenAI's research blog, Anthropic's announcements, Google DeepMind, Hugging Face, and Meta AI. Those are not blogs. They're product PR channels staffed by marketing teams. They tell you what a frontier lab shipped; they do not tell you whether it matters, how it works in practice, or who's been right about where the field is heading.
This roundup is the alternative — the independent practitioners, researchers, engineers, and writers whose work the corporate blogs end up citing, not the other way around. It draws from two sources: 14 AI/LLM-focused blogs that passed our scoring engine and live in the AwesomeBloggers directory, plus 27 hand-picked external recommendations across blogs, newsletters, and podcasts that we don't (yet) score but absolutely read. Directory entries get a numeric rank and an Awesome Score; external recommendations are labeled as such and linked directly to the source.
If you've already subscribed to The Batch and read Simon Willison's blog, the goal here is to give you the next 20 things worth reading — and to mark which ones are worth your inbox rather than just your bookmark folder. As of June 2026, this is what we recommend.
At a Glance: Top Picks by Use Case
If you only have time for one recommendation per category, this is what we'd hand to you (as of June 2026):
| You want | Top pick | Runner-up |
|---|---|---|
| Daily field reports on LLMs | Simon Willison | AINews |
| Code-and-math LLM tutorials | Sebastian Raschka / Ahead of AI | Deep Learning Focus |
| AI for non-engineers | One Useful Thing (Ethan Mollick) | The Algorithmic Bridge |
| Open-model and post-training analysis | Interconnects (Nathan Lambert) | The Gradient |
| Critical takes on AI hype | AI Snake Oil (Princeton) | Marcus on AI |
| Frontier lab interviews | Dwarkesh Podcast | Latent Space |
| Daily AI news firehose | TLDR AI | Ben's Bites |
| Visual ML explainers | Maggie Appleton | colah's blog (archive) |
| Production AI engineering | Hamel Husain | Jason Liu (jxnl.co) |
| Quantitative AI data and trends | Epoch AI | Gradient Updates |
| Weekly AI roundup, exhaustive | Don't Worry About the Vase | Last Week in AI |
| Weekly AI roundup, friendly | The Batch (Andrew Ng) | Import AI (Jack Clark) |
The detailed clusters below explain why each entry made the cut and group everything by editorial focus. Skim the table if you're recommendation-shopping; read the clusters if you want to understand the landscape.
How We Picked These
Directory entries cleared our 40-point quality bar across seven metrics (DA, freshness, speed, mobile, volume, SSL, design — see methodology). External entries didn't go through scoring — they're editorial picks based on signal density (insight per post), depth of expertise (does the author actually do the work they write about), and current activity (is the publication still publishing as of mid-2026). We deliberately excluded every corporate research blog, every VC-funded AI media outlet that reads like recycled press releases, and every AI-tool directory site that lists tools without ranking or testing them.
The six clusters that follow are not strict — many entries belong in two — but they describe what each publication is genuinely best at. Skip to whichever section matches what you actually want from your reading time.
The Field Reporters
Daily and weekly accounts of building with LLMs. These are not research surveys. They're field notes from people who shipped something this week and are reporting back on what worked and what didn't.
1. Simon Willison — Awesome Score 79/100
Django co-creator, maker of the llm command-line tool, and effectively the historical record-keeper of the LLM era. Simon publishes daily — often multiple times — and tracks every meaningful model release, tool launch, and benchmark in real time on simonwillison.net. The most consistent single source for "what just shipped and why I care." Also publishes a monthly digest newsletter for sponsors at $10/month.
2. Daniel Miessler — Awesome Score 78/100
Cybersecurity practitioner who covers AI with the same rigor he applies to threat modeling. Author of the Unsupervised Learning newsletter (530+ issues), which sits at the intersection of national security, AI policy, and prompt engineering. Stronger on AI's downstream implications than its model internals.
3. Chip Huyen — Awesome Score 42/100
Author of Designing Machine Learning Systems and the forthcoming AI Engineering. Writes on ML systems design — evals, data, deployment, the unglamorous decisions that separate working AI products from demos. Cadence has slowed in 2025–2026 (last post January 2025) but the back catalog remains the closest thing to a textbook for production ML.
Hamel Husain — Recommended (not in our directory)
hamel.dev — Practitioner blog on building, evaluating, and shipping LLM-powered products, with an emphasis on evals, error analysis, and fine-tuning. His January 2026 "LLM Evals: Everything You Need to Know" FAQ is the most-cited single reference for production eval design. His joint evals course with Shreya Shankar has trained 700+ engineers.
Jason Liu — Recommended (not in our directory)
jxnl.co — Created the Instructor Python library (the one OpenAI cited as inspiration for structured outputs) and now works on the OpenAI Codex team. His Systematically Improving RAG Applications series is the most circulated practical framework for retrieval pipelines. Also writes openly about the business of independent AI consulting.
Latent Space — Recommended (not in our directory)
latent.space — Shawn Wang (swyx) and Alessio Fanelli's newsletter and podcast, built explicitly for the role they helped define: the AI Engineer. Long interviews with the builders behind Cursor, Replit, Cognition, and frontier-lab tooling, often before anyone else publishes them. 185,000+ subscribers; also the publisher behind the AI Engineer Summit conference series.
AI Tidbits — Recommended (not in our directory)
aitidbits.ai — Sahar Mor (ex-Stripe AI) writes for people shipping LLMs in production. The "AI Coding Series" walks through Claude Code in real codebases rather than toy demos. Weekly editorial deep dives focused on accuracy, security, and cost — the things that matter when an LLM call hits production traffic.
The Research Translators
People who read the papers, run the code, and explain what's actually new. The boundary between "I taught myself this from scratch" and "I learned it from this blog" runs through this cluster.
4. Sebastian Raschka — Awesome Score 81/100
Author of Build a Large Language Model From Scratch and arguably the most accessible LLM-architecture writer working. Raschka's blog and accompanying Ahead of AI newsletter (193,000+ subscribers) translate frontier ML research into the kind of explanations engineers can actually code against. The annual state-of-the-art LLM reviews are reference material across the ML community.
5. Vicki Boykis — Awesome Score 82/100
ML engineering at the production layer — recommender systems, the unsexy operational reality of ML in big companies, and reasoned critiques of the latest hype cycles. Her "What are embeddings?" PDF is one of the most-shared single technical explainers of the LLM era.
6. Lilian Weng — Awesome Score 54/100
Ex-OpenAI VP of Safety; writes deep technical surveys on ML topics (attention, agents, hallucination, prompt engineering). Cadence is deliberately slow — roughly monthly to quarterly — but every post is referenced and bookmark-worthy. The closest modern equivalent to a Distill-style canonical survey.
7. Jay Alammar — Awesome Score 34/100
The Illustrated Transformer is how a generation of practitioners actually understood attention. Cadence has slowed (last post March 2025) but the back catalog of visual explainers — Transformer, BERT, GPT — remains the default first reading for anyone new to the architecture.
8. BAIR Blog — Awesome Score 63/100
The Berkeley AI Research lab's blog. Cutting-edge research explainers written by the grad students who did the work. Strongest cluster: robotics, RL, and emerging frontier capabilities that haven't crossed into industry blog coverage yet.
The Gradient — Recommended (not in our directory)
thegradient.pub — Nonprofit magazine founded at Stanford AI Lab in 2017; one of the few remaining venues for peer-edited, citation-heavy research explainers written by AI researchers for the AI community. The companion Perspectives on AI podcast runs deep technical interviews.
Deep (Learning) Focus — Recommended (not in our directory)
cameronrwolfe.substack.com — Cameron R. Wolfe (research at Netflix) writes topic-by-topic surveys that sit between Sebastian Raschka and arXiv summary newsletters — long enough to teach the math, short enough to finish in one sitting. Recent surveys on RLHF, mixture-of-experts, and rubric-based rewards have become widely circulated reference material.
colah's blog — Recommended (not in our directory)
colah.github.io — Christopher Olah (Anthropic co-founder, runs interpretability there). Cadence is effectively dormant on the personal blog now, but "Understanding LSTM Networks" and the older visual neural network essays remain the canonical reference for visual ML explanation. Newer work appears via Anthropic's Transformer Circuits thread (corporate, but the exception that proves the no-corporate-blogs rule).
The Open-Model and Hands-On Engineers
Code-first writing about production ML systems and the open-model ecosystem. If the previous cluster explains how the models work, this one explains how to make them work for you.
9. Eugene Yan — Awesome Score 71/100
Applied scientist at Amazon. Writes about ML systems and recommender systems with unusual depth on the operational details — evaluation, monitoring, the architecture decisions that compound over time. The "Patterns for LLM-based systems and products" post is a working reference for anyone designing an LLM application from scratch.
10. Machine Learning Mastery — Awesome Score 68/100
Practical ML tutorials from beginner to advanced. 1,900+ posts; the single largest archive of code-walking ML content on the indie web. Less original analysis than the rest of this cluster, but the depth and breadth of tutorials make it the default reference for "show me the code" learners.
11. Signals — Awesome Score 73/100
A solo founder's build log — constructing cryptographic data infrastructure for AI agents, one grep at a time. Newer than the rest of the cluster but worth following because it captures, in real time, the unglamorous reality of shipping AI infrastructure as a one-person team. The kind of post-mortem-while-it's-happening writing the field needs more of.
Interconnects — Recommended (not in our directory)
interconnects.ai — Nathan Lambert leads post-training of the OLMo and Tülu open models at Allen Institute for AI. When he writes about RLHF or the open-vs-closed model performance gap, it's coming from someone running the experiments himself. The de facto source for open-weights model analysis. Also writing the first textbook on RLHF.
The Newsletter Stack
Email-first publications worth a permanent inbox slot. Most are also blogs, but the format is the newsletter — usually weekly, often Sunday or mid-week. Organized by what you'd actually subscribe to them for.
Import AI — Recommended (not in our directory)
jack-clark.net — Jack Clark, Anthropic co-founder and policy lead. Weekly, 458+ issues. The only major AI newsletter that pairs rigorous paper analysis with original short fiction imagining the technology's future ("Tech Tales"). Crossed 100,000 subscribers in 2026. The newsletter where a frontier-lab co-founder publicly works through his own probability estimates for autonomous AI R&D.
The Batch — Recommended (not in our directory)
deeplearning.ai/the-batch — Weekly curated AI news with a personal letter from Andrew Ng at the top of every issue. Ng's framing is the differentiator — you get a Stanford ML legend's take on the week's news in plain language. The best on-ramp for engineers who want signal without the X/Twitter noise.
Last Week in AI — Recommended (not in our directory)
lastweekin.ai — Andrey Kurenkov (Stanford AI PhD) and co-hosts cover roughly 30–50 stories per week with brief technical commentary on each. Broader scope than any practitioner-written newsletter. The "I read it so you don't have to" subscription for AI generalists. 249,000+ subscribers; podcast goes deeper than the newsletter.
Ahead of AI — Recommended (not in our directory)
magazine.sebastianraschka.com — Sebastian Raschka's newsletter (separate from his blog). The most code-and-math-dense publication on this list. 193,000+ subscribers. Best subscription if you want to actually understand LLM architectures rather than just consume the news about them.
AINews — Recommended (not in our directory)
news.smol.ai — Daily weekday recap of top discussions across AI Discords, AI subreddits, and AI Twitter, produced 99% by customizable research agents with human editorial sweep on top. Endorsed by Andrej Karpathy as "best AI newsletter atm" and Soumith Chintala as the "highest-leverage 45 minutes I spend everyday." 80,000+ subscribers. The only newsletter that systematically summarizes community channels rather than press releases.
Don't Worry About the Vase — Recommended (not in our directory)
thezvi.substack.com — Zvi Mowshowitz's exhaustive weekly AI roundups ("AI #149+") plus long-form analysis. Often 15,000+ words. The longest, most detailed single-newsletter coverage of the AI beat. The rationalist-community reference point for AI safety debates. Reads everything so you can read just him.
Gradient Updates — Recommended (not in our directory)
epoch.ai/gradient-updates — Weekly single-topic commentary from the Epoch AI research team (Ege Erdil and others). One question per week — "How close is AI to taking my job?", "What do frontier lab roadmaps imply?" — answered with data, not vibes. Epoch produces the dataset on AI compute and training trends that everyone else cites.
Chain of Thought (Every) — Recommended (not in our directory)
every.to/chain-of-thought — Dan Shipper writes the rare AI newsletter that treats the technology as a humanities subject, not just a benchmark race. Paired with the AI & I podcast where he uses ChatGPT or Claude live with guests. Best for operators and creatives, not engineers. Part of the Every network (180,000+ subscribers).
The Algorithmic Bridge — Recommended (not in our directory)
thealgorithmicbridge.com — Alberto Romero translates frontier AI research into essays for non-technical professionals without losing rigor. 44,000+ subscribers; ranked #12 Rising in Technology on Substack. The clearest writer in the "AI for educated non-engineers" category.
One Useful Thing — Recommended (not in our directory)
oneusefulthing.org — Wharton professor Ethan Mollick runs hands-on experiments with frontier models and reports what works. 441,000+ subscribers — the largest serious AI newsletter outside the daily aggregators. The newsletter most likely to be cited in a corporate training deck, and most likely to be read by C-suite executives trying to make sense of AI.
High-Volume Daily Aggregators
Three giant prosumer newsletters cover the AI news firehose for a general professional audience. Editorial voice is weaker than the practitioner picks above, but the distribution is real and the picks are decent.
TLDR AI — Recommended (not in our directory)
tldr.tech/ai — Daily 5-minute AI news digest written for ML engineers, infra teams, and technical PMs. Heavier on research and infrastructure than the consumer-focused alternatives. Part of the TLDR network; 1,250,000+ subscribers.
Ben's Bites — Recommended (not in our directory)
bensbites.com — Ben Tossell built this for founders and pre-seed investors, and that's still the audience. One of the few daily AI newsletters that consistently picks apart wrapper economics, frontier-lab moats, and YC AI startup launches without parroting press-release framing. 163,000+ subscribers.
The Rundown AI — Recommended (not in our directory)
therundown.ai — The largest AI newsletter on the internet at 2,000,000+ subscribers. Editorial voice closer to a tech-news daily than to practitioners — but the default first AI newsletter most professionals subscribe to, and Rowan Cheung's interviews with frontier-lab leaders occasionally get genuine scoops.
The Skeptics and Critics
The counterweight cluster. If you only read frontier-lab newsletters and practitioner blogs, you will internalize a specific worldview about where AI is heading. These four publications exist to push back on it.
AI Snake Oil — Recommended (not in our directory)
aisnakeoil.com — Arvind Narayanan and Sayash Kapoor (Princeton). Authors of the 2024 Princeton University Press book of the same name. Their "AI as Normal Technology" essay has become the leading intellectual counterweight to both scaling-maximalist and AI-doom narratives. 76,000+ subscribers; both authors made TIME100 AI. Read this newsletter specifically to test your assumptions about LLM capabilities and limits.
Marcus on AI — Recommended (not in our directory)
garymarcus.substack.com — Gary Marcus is the highest-profile dissenting voice on LLM capability claims and a recurring US Senate witness on AI policy. 106,000+ subscribers. The blog so influential that an independent project tracks his predictions systematically. Read with awareness of the author's track record (some calls validated, some not), but don't dismiss the framework just because the specific predictions are contested.
Planned Obsolescence — Recommended (not in our directory)
planned-obsolescence.org — Ajeya Cotra (Open Philanthropy's lead AI grantmaker; author of the biological anchors timelines report) and Kelsey Piper (Vox Future Perfect's lead AI journalist). The blog pairs the leading alignment grantmaker and the leading alignment journalist — a perspective combination nothing else offers. Newer than the others in this cluster but already a key alignment-discourse venue.
Frontier Interviews and Quantitative Sources
Two niches that deserve their own callout: the long-form interview format and the quantitative-data-on-AI format. Both are scarce in the indie space.
Dwarkesh Podcast — Recommended (not in our directory)
dwarkesh.com — Long-form, deeply researched interviews with frontier AI lab leaders, scaling researchers, and historians. Guests include Dario Amodei, Ilya Sutskever, Demis Hassabis, Satya Nadella, and Jensen Huang. 82,000+ subscribers. Currently the highest-signal interview venue in AI. Companion essays and transcripts on the site.
Gwern.net — Recommended (not in our directory)
gwern.net — Gwern Branwen's decade-spanning essay site on AI scaling, statistics, self-experimentation, and the long tail of obscure technical research. His "The Scaling Hypothesis" (and 2022 revisit) was one of the earliest and most influential public statements of the thesis that drove the GPT-3/4 era. Uses unique progressive-disclosure web design; pages continue to update for years after first publication.
Epoch AI — Recommended (not in our directory)
epoch.ai/blog — Quantitative research blog and data hub tracking training compute, inference costs, dataset sizes, and AI hardware capacity over time. The canonical source for AI scaling data — their "compute trends across three eras of ML" framework and ongoing trackers are cited in nearly every serious AI policy and economics paper.
The Visual Thinkers
A short cluster but a necessary one. The visual-essay tradition that produced Distill is alive, just distributed across fewer outlets.
12. Maggie Appleton — Awesome Score 70/100
Digital garden with visual essays on programming, AI, and design. Recent series on "anthropomorphic" and "ambient" interfaces for AI products is some of the sharpest thinking on AI UX in public. Cadence is slower than her newsletter-writing peers (last post April 2026) but the depth is the point.
Distill — Historically essential, no longer publishing
distill.pub — Last post September 2021. The interactive ML visualization journal that set the standard everyone else is now trying to meet. The archive remains a working reference; many of the foundational visual explainers of attention, deep learning, and feature visualization still live here.
Foundational Reading
One callout for a blog that's too sporadic to rank but too important to omit.
Andrej Karpathy — Awesome Score 53/100
karpathy.github.io — Former Tesla AI director, ex-OpenAI founding member, and creator of the most widely watched "build a GPT from scratch" video series on YouTube. Blog posts come quarterly at best, but each is treated as canonical reading the moment it lands. Pair the blog with the YouTube channel for the full effect.
What This List Excludes (and Why)
We deliberately left out three categories that show up in almost every other "best AI blogs" list:
Corporate research blogs (OpenAI, Anthropic, Google DeepMind, Meta AI, Nvidia). These are product PR channels with marketing budgets. The work is often excellent, but the framing is always "look what we shipped." Read them as primary sources for what a lab claims, not as analysis. The single exception we noted — Anthropic's Transformer Circuits thread — is a corporate-published research venue with genuine peer-reviewed depth.
Venture-funded AI media outlets (Towards Data Science, KDnuggets, Marktechpost, Analytics Vidhya, Built In's AI coverage). High publishing volume, weak editorial signal-to-noise, increasingly written by anonymous contractors for SEO. The opposite of what this list is trying to find.
AI tool directories (There's an AI for That, Futurepedia, AI Tool Hunt). These are listing sites, not blogs. Useful for tool discovery; not useful for thinking.
Our Source Data
The 14 directory blogs come from the Tech & AI category of AwesomeBloggers — 59 approved indie tech blogs as of June 2026. Awesome Scores reflect our seven-metric scoring engine, refreshed weekly. The AI-specific subset above was selected by reading each blog's stated focus and recent posts, not by automated keyword matching.
External recommendations were curated by hand. Each was verified for current activity (active or actively updating within the last 60 days), specific differentiator (what makes it worth reading rather than substitutable with a corporate blog), and depth of expertise (does the author actually do the work they're writing about). We did not include blogs we hadn't read.
The average domain authority of indie blogs in our directory is OPR 2.97 — the AI-specific subset above clusters slightly higher at OPR 4 because tech-audience link graphs compound differently than other niches. The publishing frequency study found that Tech & AI blogs publish weekly at 5% — below food and personal finance — but the consistent monthly cadence (44% within 30 days) makes the cluster reliable to subscribe to. Anecdotally, AI/LLM writers we read lean more toward newsletter format than the indie writers in other niches; we'd want a separate study to put a number on that.
Frequently Asked Questions
Q: What is the best AI newsletter for engineers in 2026?
A: For working AI engineers shipping LLM products, the highest-signal subscription is Latent Space — built explicitly for the AI Engineer role, with podcast interviews of the people building Cursor, Replit, Cognition, and frontier-lab tooling. Pair it with Ahead of AI (Sebastian Raschka) for code-and-math-dense architecture explainers and Interconnects (Nathan Lambert) for open-model and post-training analysis. Those three cover most of the working-engineer space.
Q: What's the best AI newsletter for non-engineers?
A: One Useful Thing by Ethan Mollick (Wharton). Hands-on experiments with frontier models, framed for educators, managers, and operators rather than ML researchers. 441,000+ subscribers as of mid-2026 — the largest serious AI newsletter outside the daily aggregators. The Algorithmic Bridge (Alberto Romero) is the runner-up if you want longer essayistic writing.
Q: Are corporate AI lab blogs (OpenAI, Anthropic, DeepMind) worth subscribing to?
A: As primary sources for what a lab is claiming about its own models, yes. As analysis of what those models actually mean, no — they're written by marketing teams whose job is to make the lab look good. Read them for the announcement; read the writers in this list for the analysis.
Q: Which AI blog should I read first if I'm new to the space?
A: For ML fundamentals, start with The Illustrated Transformer (Jay Alammar) and Sebastian Raschka's series on building an LLM from scratch. For "what's happening in AI right now," subscribe to The Batch (Andrew Ng's letter is the best on-ramp) and One Useful Thing (Ethan Mollick). For the dissenting view, add AI Snake Oil.
Q: How often do AI blogs publish?
A: The AI/LLM cluster skews more toward newsletter format and less toward weekly blog cadence than other niches. Daily publishers are rare (Simon Willison and AINews are the main ones). Most serious writers publish weekly to monthly. Per our publishing frequency data study, 5% of Tech & AI blogs in our directory publish weekly and 44% publish within 30 days — slightly below the cross-category average. The trade-off is depth: AI newsletters tend to be longer and more analytical per post.
Q: Are these all free?
A: Most are free with optional paid tiers. Notable paid-only or paywalled: Stratechery (not on this list — see exclusions), much of The Sequence (excluded for this reason). Simon Willison's monthly newsletter is sponsor-supported at $10/month but the blog itself is free. Latent Space, Interconnects, Ahead of AI, and Don't Worry About the Vase all have substantial free tiers.
Q: Why isn't [X corporate blog] on this list?
A: We deliberately excluded corporate research blogs (OpenAI, Anthropic, Google DeepMind, Meta AI, Nvidia) because they're product PR channels rather than independent analysis. The one exception — Anthropic's Transformer Circuits thread — is genuinely a research venue rather than a marketing channel. We also excluded venture-funded AI media outlets (Towards Data Science, KDnuggets, Marktechpost, Analytics Vidhya) for editorial signal-to-noise reasons. AI tool directories (There's an AI for That, Futurepedia) are also out — they're listing sites, not blogs.
Q: How was the directory ranking determined?
A: The 14 numbered entries are blogs in our Tech & AI category that focus primarily on AI/LLM/ML topics, ranked by Awesome Score from our seven-metric scoring engine. External recommendations are not ranked numerically — they're editorial picks, organized by cluster.
The dataset of scored indie blogs behind this article grows weekly as new submissions come in. If you write an AI/LLM blog that should be in our Tech & AI directory, submit it here — scoring runs in real time. If you think we missed an essential external recommendation for the next refresh of this article, the same submission form works.