Feature a build or sponsor the directory. DM @pranavmore69
Long threads people actually finish. Essays, deep dives, and the occasional story that sticks.
35 threads · updated 2026-09-21
35 of 35 threads
@AIatMeta
Official deep dive. Secure VM, Sentinel on the same box, prompt-injection quarantine, public bug bounty. First-party, long, and the reason they claim the agent can hold your context.
Sep 21, 2026
@MetaNewsroom
Meta's product note. Personal agent, Muse Spark 1.3, Secure VM, Stripe Link for purchases, free for most of what people need.
Sep 21, 2026
@svpino
Santiago on the Agents CLI. Build, deploy, and watch a small multi-agent system without the usual harness ceremony.
Sep 16, 2026
@dotey
Baoyu's Tencent talk. Find demand on the model capability line, kill bad ideas with prototypes, then redesign so the agent verifies itself.
Sep 2, 2026
@poteto
Lauren on verification as infrastructure. Agents close their own loop, Feature Maps beat worktrees, and volume only works if quality holds.
Sep 1, 2026
@davidgomes
David Gomes on harness-managed git worktrees versus letting the agent own the tree. The second one tends to become abstract art.
Aug 29, 2026
@GaryMarcus
Gary Marcus, with Zack Korman, on what OpenAI should have done, not just what the model did.
Aug 29, 2026
@finn_fergus
Fergus Finn follows an RTX 4090 STG.E from the warp through L1, L2, dirty lines, write-back, and DRAM. The store finishes in 6 cycles. The data gets to DRAM later.
Aug 29, 2026
@JohnNosta
John Nosta on AI-detection theater. The question is not whether the model wrote the essay. It is whether the student did the thinking.
Aug 28, 2026
@danieldines
UiPath founder Daniel Dines on the limit that actually matters: AI reasons, but it does not learn on the job. A hire absorbs what nobody wrote down. The model forgets overnight.
Aug 27, 2026
@danwilliamsphil
Sussex philosopher Dan Williams on why the hard questions cut across disciplines, and why this is a golden age for polymaths.
Aug 27, 2026
@TaylorLorenz
Taylor Lorenz on Claudish, the dialect people pick up after too many hours with coding agents.
Aug 27, 2026
@paulg
Paul Graham's new essay: teach them to build, make starting a company feel possible, and get out of the way of their own projects. Not an entrepreneurship major.
Aug 26, 2026
@maiamindel
Maia Mindel on Daron Acemoglu, settler mortality, and whether the pile-on was character assassination or an actual argument.
Aug 25, 2026
@Afinetheorem
Kevin Bryan on what students still have to learn without a model, what to stop assigning because it is cheatable, and how to grade when incentives actually matter.
Aug 24, 2026
@ProfBuehlerMIT
MIT's Markus Buehler ran a team of Grok Bots from photos to a physics lab to a Bambu printer. The whole scientific loop, in one night.
Aug 22, 2026
@ArianeGroup
ArianeGroup on the engineering of fairing release: too early and too late both cost you. A short, actual rocket thread.
Aug 21, 2026
@startupideaspod
Greg Isenberg's whole company runs on skills in a repo. Add a marketplace URL in Claude Code, split by department, auto-update.
Aug 20, 2026
@japan_nobunaga
A 98-year-old Navy sailor died with no family. One funeral-home post. A Japanese writer stood outside because the church was full.
Aug 20, 2026
@simonw
Simon Willison asked Claude Code to try smolvm as a sandbox. No /dev/kvm, so it wrote a GitHub Actions workflow and pushed it. Unasked.
Aug 20, 2026
@oprydai
Math, physics, chemistry, biology, CS, economics, then engineering. A robot is physics plus computation plus control.
Aug 19, 2026
@dabit3
Nader Dabit on high-ROI agent workflows: four you can stand up in a minute, and forty more once the language is clear.
Aug 19, 2026
@patio11
Patrick McKenzie's eight-year-old gets stung, asks why, then insists the bee deserves honors. They bury the bee.
Aug 18, 2026
@simonw
Simon Willison writes up the Black Hat talk: what OpenAI says happened, in order. Pretty wild.
Aug 7, 2026
@arpit_bhayani
Writing code used to be expensive, so process protected engineering time. Arpit Bhayani on what breaks when that assumption dies.
Aug 6, 2026
@GergelyOrosz
A Head of Engineering spent a week in SF. The memo: the impressive teams were still searching for a business. You already have one.
Jul 7, 2026
@adxtyahq
Mental model, production loop, and why the hardest part of AI is no longer the model. Video plus a write-up.
Jul 2, 2026
@levelsio
Pieter Levels on Remote OK, Nomads, Photo AI, and the normal-shaped revenue curve. AI products peak faster.
Jun 30, 2026
@karpathy
Don't emulate one PhD student. Emulate a research community. Git is almost, but not quite, the right abstraction.
Mar 8, 2026
@dhh
DHH's polemic on why small teams shatter their own superpower — shared context — and call the debris architecture.
Dec 10, 2025
@karpathy
A full guide to adding a capability to a tiny LLM: synthetic tasks, tokenization traps, and why a honeybee-sized model needs the data over-represented.
Oct 24, 2025
@karpathy
AGI as a decade of agents, animals vs ghosts, RL as a straw, cognitive core, and why he does not want 1,000 lines of unsupervised code.
Oct 18, 2025
@levelsio
If most goods go to zero, Pieter's list is land, energy, fabs, luxury, human connection, security, and biotech.
Sep 5, 2025
@swyx
Six elements of the job, why agents are ChatGPT's path to a billion MAU, and why work on it now.
Mar 24, 2025
@swyx
Swyx's 2023 essay that named the job. Code that orchestrates models, not just trains them.
Jun 30, 2023
Meet Muse, your personal AI agent. Muse doesn’t just answer questions, it actually does the work across the apps you already use. It helps you stay on top of things, takes tasks off your plate, and turns long-term goals into action plans. about.fb.com/news/2026/09/i…
The Agents CLI is the closest thing to magic you'll ever use when building agents. I wrote an article to show you how to use it. It will take you 30 minutes to build, deploy, and start monitoring a small multi-agent system. x.com/i/article/2098…
I'm writing a guide to pstack! Here's part one. x.com/i/article/2094…
Now live (and free) at Marcus on AI: takeways on the OpenAI / Hugging Face attack, coauthored with @ZackKorman, focusing not so much on what the AI did as on what OpenAI should have done. garymarcus.substack.com/p/5-lessons-fr…
New post: continuing our reverse engineering of the memory hierarchy of an RTX 4090 This time we follow a STG instruction down from the warps to DRAM. Cache replacement policies, dirty lines & write-back, visibility, and coherence. blog.doubleword.ai/what-happens-w…
🚨The problem with AI detection is that it completely misses the point. The real question isn't whether AI wrote the essay, it's whether the student did the thinking. 🔥The Educational Artifact Is No Longer Enough psychologytoday.com/us/blog/the-di… #AI #education #learning Show more
My book is out today: The Work That Remains — Human Judgment, AI, and the Architecture of the Next Enterprise. I wrote it because most of what executives hear about AI comes from one of two rooms — one in panic, one in euphoria. Neither helps you decide anything. What Show more
New essay! A big-picture, speculative article on how most of the huge questions about the coming AI revolution aren’t really about AI in a narrow, technical sense - and why it will be a golden age for polymaths, philosophers, and moderate epistemic despair. Link: Show more
I wrote about the rise of Claudish and how people who spend a lot of time with coding agents are able to speak an entirely new dialect usermag.co/p/vibe-coders-…
How Universities Should Prepare Founders: paulgraham.com/prepare.html
New post: why was everyone freaking out about Daron Acemoglu? Was it all just character assasination by AI shills and Jeffrey Epstein’s friends? Is he actually The Real Deal? And what’s the deal with settler mortality, anyways? someunpleasant.substack.com/p/extraordinar…
Fall terms coming up. How should we modify university teaching to both use AI effectively and ensure students actually learn/don't just cheat their way through? I wrote eight rules I've been using; hopefully useful for you as well. 1/3
🚀 How is choosing the right moment to release the fairing an engineering challenge? Because separating it too early... or too late... both come with consequences. 🧵
My whole company runs on skills I bundled into a GitHub repo. Here's how to build the same thing for your team: - Put your skills in a GitHub repo - Get Claude to write the couple of JSON files that make it readable as a plugin - In Claude Code, type /plugin, go to Show more
x.com/i/article/2090…
An American I barely knew texted me at nine at night. "You free Saturday?" I asked what for. He said, "A funeral." I asked whose. He said, "No idea." This is how I learned about John. Ninety-eight years old. Navy, World War Two. Served on the USS Houston. Visited twenty-seven Show more
I had Claude Code for web experiment with smolvm as a code execution sandbox Fable 5 spotted that its environment couldn't run that (no /dev/kvm)... so, without asking me first, it wrote a GitHub Actions workflow to run the experiments and pushed that directly to GitHub instead!
to thrive in the future, understand the fabric of reality from an engineering perspective. don’t just learn tools. learn the layers underneath them. • math → the language of patterns, structure, change, uncertainty • physics → the rules matter and energy must obey • Show more
High ROI skill: driving 24/7 app maintenance + engineering automations, and a key component of a real software factory. Implementing these workflows is simpler than ever, you just need clear language. Here are 4 workflows and 40+ more you can build in ~1 minute each 🧵
Liam, 8; I got stung by a bee! *moans* Me: I’m sorry. Liam: Why do bees sting? Me: In defense of themselves or their hive. Liam: I was nowhere near a hive. Me: He might have made a mistake. You’re ten thousand times bigger than him and so maybe he was terrified. Liam, suddenly…
Thanks to the video from the Black Hat security conference of OpenAI's presentation about "The Hugging Face Incident" we now have a detailed timeline of what happened from OpenAI's perspective - I wrote up the details here, it's pretty wild simonwillison.net/2026/Aug/7/ope…
AI did not just change how we write code. It changed what the bottleneck in software engineering is. For the last two decades, most engineering processes were designed around one assumption - writing code was expensive, so we built process around protecting engineering time. Show more
From a Head of Engineering who spent a week in SF, meeting a bunch of AI startups: "When I was back, I wrote a memo about my impressions. The #1 was how these amazing startups I met: they mostly did not have a business. But we do. So be VERY careful in copying what they do. Show more
loop engineering has been all over my timeline lately, so i finally sat down and made a proper explanation of how it actually works covers the mental model, the production loop, and why the hardest part of AI isn't the model anymore. also wrote a blog if you'd rather read: Show more
Here's some revenue stats that might be interesting: 1) My YouTube channel 2) Remote OK 3) Nomads 4) Photo AI So I think it's just the natural life cycle of a product, with Remote OK it peaked 6 years after starting, with Nomads it peake after 8 years! They all follow a Show more
The next step for autoresearch is that it has to be asynchronously massively collaborative for agents (think: SETI@home style). The goal is not to emulate a single PhD student, it's to emulate a research community of them. Current code synchronously grows a single thread of Show more
Microservices is the software industry’s most successful confidence scam. It convinces small teams that they are “thinking big” while systematically destroying their ability to move at all. It flatters ambition by weaponizing insecurity: if you’re not running a constellation of Show more
Last night I taught nanochat d32 how to count 'r' in strawberry (or similar variations). I thought this would be a good/fun example of how to add capabilities to nanochat and I wrote up a full guide here: github.com/karpathy/nanoc… This is done via a new synthetic task Show more
My pleasure to come on Dwarkesh last week, I thought the questions and conversation were really good. I re-watched the pod just now too. First of all, yes I know, and I'm sorry that I speak so fast :). It's to my detriment because sometimes my speaking thread out-executes my Show more
The @karpathy interview 0:00:00 – AGI is still a decade away 0:30:33 – LLM cognitive deficits 0:40:53 – RL is terrible 0:50:26 – How do humans learn? 1:07:13 – AGI will blend into 2% GDP growth 1:18:24 – ASI 1:33:38 – Evolution of intelligence & culture 1:43:43 - Why self
Great thread about investment post-AGI I've been thinking a lot about it recently and switching my investments up a bit Post-AGI there's a prediction the prices of almost everything will go to close to zero Think for example the food you eat, if it can be farmed fully Show more
Recently, I've noticed people making a big deal about how we haven't yet seen massive disruption to job markets and knowledge work from AI, and so they're starting to doubt that it will happen soon. And investors are wondering if they can just sort of ignore it for a while.
🆕 talk + essay: Agent Engineering latent.space/p/agent Why we went all in on Agents @aiDotEngineer Defining Agents (thanks to @simonw) The Six Elements of Agent Engineering Why Agents are ChatGPT's path to 1B MAU Why work on Agent Engineering Now live on @latentspacepod! Show more
🆕 Essay: The Rise of the AI Engineer latent.space/p/ai-engineer Keeping up on AI is becoming a full time job. Let's get together and define it.
