Turns out OpenAI found out what we did over a month ago, Kimi2.5 is good at coding and The Zero-Human Company shows how we used it as Co-CTOs. OpenAI is just putting a wrapper on it. In my garage me and Mr. @Grok CEO just build what do they do with billions? https://t.co/bjTyqRNX28 https://t.co/1oaFT2fzZV
— Brian Roemmele (@BrianRoemmele) Mar 20, 2026
We are dangerously close to putting Codex in autonomous loops where it picks up tickets, tests it's own changes via Playwright, and records and uploads verification mp4s to PRs. If you've never asked Codex to test your app, do try it! https://t.co/hfcqspi26t https://t.co/eUtjbBL116
— Nick (@nickbaumann_) Mar 20, 2026
🚨 China has released an AI employee that runs 100% locally. It can do research, code, build websites, create slide decks, and generate videos.. all by itself. And it comes with its own computer. 100% Open Source. https://t.co/KuhKCteBYA
— Alvaro Cintas (@dr_cintas) Mar 21, 2026
Vercel, which helps developers host web apps and AI agents, says its run-rate GAAP revenue hit $340M at the end of February, up 86% YoY, amid the AI coding boom (@richardjnieva / Forbes) https://t.co/gy5ftaFcoB https://t.co/LjBVULuT3F 📥 Send tips! https://t.co/wlNZvXuhJs
— Techmeme (@Techmeme) Mar 21, 2026
gpt-5.4 for frontend: https://t.co/LbdQKyEhoU
— Greg Brockman (@gdb) Mar 21, 2026
Sequoia’s Doug Leone explains what made Drew Houston’s pitch for Dropbox great Dropbox raised a $1.2 million seed round at a $5 million valuation from Sequoia in 2007. Eleven years later, the company went public at a valuation of $9 billion. In hindsight it’s a home run, but in https://t.co/PGekn2O63V
— Startup Archive (@StartupArchive_) Mar 21, 2026
The new @karpathy podcast with @NoPriorsPod is a must-watch. I turned the most surprising insights & anecdotes into a webpage. For each, you can easily jump to the timetamp in the video where it occurred. Made with @AnyGenIO (the most underrated AI tool out there) Link below! https://t.co/1WXo7yZmRE
— Zara Zhang (@zarazhangrui) Mar 20, 2026
Got served this listing for a house in the Hollywood Hills and it’s so gorgeous that it almost makes me want to live in a completely car-dependent area in LA Also - how is this so cheap? https://t.co/6uo0iNL67E https://t.co/DzZZQBTgAv
— Sheel Mohnot (@pitdesi) Mar 20, 2026
Markdown is code. I am in a branch where I am modifying GStack so that it can automatically do Codex code and plan reviews as a default setting. Of course I have the skill vendored in the repo so my workspace uses whatever is local. So as I was actually running /ship it ran https://t.co/YpyYtT472r
— Garry Tan (@garrytan) Mar 20, 2026
As we transition to the agentic enterprise. Stil work to be done to fully model the value of intelligence manufacturing plants (AI Factories). https://t.co/HbYZCQFhhC
— Ben Bajarin (@BenBajarin) Mar 20, 2026
Andrej Karpathy just went ~66 mins on No Priors Podcast with Sarah Guo about code agents, AutoResearch, and what happens when humans become the bottleneck in their own systems. The clearest thinking I have heard on what just changed in December 2025 and why everything feels https://t.co/dJLWB4CIeH
— Anish Moonka (@AnishA_Moonka) Mar 20, 2026
this guy's last article got 2,700,000+ views he just dropped his blueprint to the best Openclaw setup these 12 tips will make your system into a beast my favourite are >treating main agent as manager >seperating chats by topics >scheduling overnight jobs >multi model https://t.co/PEAupUIija https://t.co/XyNqRGPo8m
— Avid (@Av1dlive) Mar 20, 2026
The 3 files in this repo are the most elegant product spec ever shipped. program.md is the PRD. Plain English research directions, constraints, and goals. The human writes this. https://t.co/VeJUoqJtVn is the engineering team. The agent reads the spec, forms a hypothesis, edits https://t.co/i9xERLdTiX
— Aakash Gupta (@aakashgupta) Mar 20, 2026
I think I finally figured out why OpenClaw is amazing and took off like wild fire and why Peter is a genius, as Altman called him. And it's actually a different way of looking at it. It's not a DeepSeek moment for agents. It's a Napster moment. And just like Napster it will https://t.co/Dk7c46Usmi
— Daniel Jeffries (@Dan_Jeffries1) Mar 20, 2026
I'm on a journey to build a software factory, and this is my open source project gift to you. https://t.co/VPvWDzV5c0
— Garry Tan (@garrytan) Mar 20, 2026
Brian Chesky wrote a great piece on how to slow down time that is relevant to this https://t.co/yDhuJJzoMB https://t.co/hZNOBPIHR4
— BOSS (@thebeautyofsaas) Mar 19, 2026
Andrej Karpathy left an AI agent running for two days. It made 700 changes to his code. Found 20 improvements he'd missed over two decades of manual work. Cut his benchmark by 11%. The whole thing is 630 lines of Python code and runs on a single GPU. I spent a week digging https://t.co/HpDjqW0oXr
— Anish Moonka (@AnishA_Moonka) Mar 19, 2026
Worth playing around with https://t.co/BafudS46GI if you haven't. Very slick experience.
— Patrick Collison (@patrickc) Mar 19, 2026
Three months ago, the consensus was that Cursor was cooked. Claude Code crossed $2.5B in run-rate revenue. Google paid $2.4B for Windsurf’s IP and poached its leadership into DeepMind. OpenAI acquired Astral, the team behind Python’s uv package manager, to feed Codex. Viral https://t.co/DDIZifIpts https://t.co/jkFpeWhZJt
— Aakash Gupta (@aakashgupta) Mar 19, 2026
My latest conversation with the always incredible @benedictevans: OpenAI’s moat problem, the rise of ephemeral and improvised sofware, OpenClaw & agents. 00:00 Intro 01:06 OpenAI's Focus Shift 03:12 ChatGPT usage: a "mile wide, inch deep" 09:03 Why better models do not solve https://t.co/LJHQaZ2nlq
— Matt Turck (@mattturck) Mar 19, 2026
Andrej Karpathy released autoresearch a few days ago, but I don't have any ML projects to try it on so I started to use it to tune some new AI skills with autoresearch all day in the background, I layered in the evals-skills from Hamel (https://t.co/pdQQsHi0oH) into the workflow
— George from 🕹prodmgmt.world (@nurijanian) Mar 18, 2026
i really like the way @sainingxie articulates his thoughts and builds elegant analogies around them. i’d strongly recommend watching the section of the podcast where he talks about research taste. one idea that stood out to me was his view that good research taste means being https://t.co/GgTMYNUOLD
— λux (@novasarc01) Mar 17, 2026
https://t.co/sdI6uMfLGs
— marv1nnnnn (@marv1nnnnn1) Mar 18, 2026
Been saying this for a while. SF is incredible for building, but the consensus problem is real. The best founders I meet are the ones closest to their customers, not closest to other startups. Wrote more on this below https://t.co/sdBU1yxtMl https://t.co/yV2tFKTgeK
— Michael Bloch (@michaelxbloch) Mar 17, 2026
big fan of langchain's technical comms. they have consistently taught me something I didn't know I needed to understand the pattern in OpenSWE is much broader than an internal SWE agent btw adjacent usecases will look more like “knowledge work in a sandbox” than software https://t.co/LUtctasH1E https://t.co/hvpiEmrrO9
— angela (@probablyangg) Mar 18, 2026
This is an awesome write up on harness engineering from @rohit4verse AI is not wrong, you just have not built the harness correctly https://t.co/rXvljNjEBn
— Aparna Dhinakaran (@aparnadhinak) Mar 18, 2026
Fantastic read on the current state and possible future of Agent Harnesses. https://t.co/Caplp7Mv81
— Ajay Kulkarni (@acoustik) Mar 17, 2026
I heard from someone that the primary reason OpenAI did the Thrive Holdings deal was to train FDEs, and the second reason was this: entering an open field of previously inaccessible training environments. These go hand in hand. In order for an FDE to automate a workflow (not https://t.co/xNoPBCy47f
— abhijay (@abhijaymrana) Mar 17, 2026
Here's the handout for a three hour workshop I presented at the NICAR data journalism conference on using coding agents (Codex CLI, Claude Code etc) for data exploration, visualization and analysis https://t.co/zriTHBGiOP
— Simon Willison (@simonw) Mar 16, 2026
As always, solid interview by @benthompson with Jensen. Lots of nuggets, but one key for me is agentic CPUs requiring the best single core performance. Seems likely agentic CPUs will require a different architectural approach. https://t.co/uovoNuv1fg
— Ben Bajarin (@BenBajarin) Mar 17, 2026
Today I'm releasing my entire newsletter archive (350+ posts) and all podcast transcripts (300+ episodes) as AI-friendly Markdown files. Plus an MCP server and GitHub repo. A few months ago I shared my podcast transcripts on a whim, and y'all built the most amazing things—an RPG https://t.co/3jTUFblaPo
— Lenny Rachitsky (@lennysan) Mar 17, 2026
New innovation: Just as at a well-run startup, there are CEO, engineering manager and design reviews, GStack now helps you keep track of what reviews are run, figures out what is appropriate (e.g. CEO doesn't have to look at infra bug fixes) and just does the smart thing https://t.co/epESEX6VLT
— Garry Tan (@garrytan) Mar 17, 2026
https://t.co/MjcCCRKXtD
— Alfred Lin (@Alfred_Lin) Mar 17, 2026
感觉 Autoresearch 会引发异一场行业变革： Autoresearch 就是一套规范。它之所以有效，是因为它强制了两件事： 纪律： 每次只改一个变量。先有假设再做实验。实验后确认或否定。这听起来显而易见，但没有这个结构的 Agent https://t.co/bWY4xVWwG1
— Tim✨ (@timyangnet) Mar 16, 2026
https://t.co/0lQ8NXM4M3
— DAIR.AI (@dair_ai) Mar 15, 2026
We built a system where Claude knows our entire company before I type a word. Three operating companies. 50+ properties. Full context on every session. Three tools. Any small business can build this. Most business owners use AI the same way every time. Open Claude. Re-explain https://t.co/QNVlq6UDzo
— Barrett Linburg (@DallasAptGP) Mar 16, 2026
Agents Over Bubbles Agents are fundamentally changing the shape of demand for compute, both in terms of how they work and in terms of who will use them. They're so compelling that I no longer believe we're in a bubble. https://t.co/LdPu3L37Vl
— Stratechery (@stratechery) Mar 16, 2026
On the back of the report on Anthropic, publishing our full growth thesis on OpenAI. @DiligenceStack OpenAI: Three Engines, One Platform https://t.co/9inLX4EVvr
— Ben Bajarin (@BenBajarin) Mar 16, 2026
Top tier explanation for non-tech newbies. Main Agents vs Subagents. (this helped me a lot) https://t.co/QnUu7AUX8F
— Vadim (@VadimStrizheus) Mar 15, 2026
Just added the 12th chapter to my Agentic Engineering Patterns guide, but it's the first one in the sequence: I figured it was time to try and answer the obvious question, "What is agentic engineering?" https://t.co/hdC7affm1E
— Simon Willison (@simonw) Mar 15, 2026
🚨Breaking: Someone just open sourced a knowledge graph engine for your codebase and it's terrifying how good it is. It's called GitNexus. And it's not a documentation tool. It's a full code intelligence layer that maps every dependency, call chain, and execution flow in your https://t.co/Jv9PFzNVfm
— Sukh Sroay (@sukh_saroy) Mar 15, 2026
have switched to https://t.co/QvK3d4AZoG as my agent harness. @badlogicgames has built something very useful. what do folks like as an extension to manage multiple agents at the same time. something that would - show me the status of various agents - show me what is stuck -
— Sriram Krishnan (@sriramk) Mar 15, 2026
Was amazing to sit down with @bgurley to talk about AI, society, politics and what it means when founder engineers can do the work of 2000 people https://t.co/Jry3KFj3ae
— Garry Tan (@garrytan) Mar 15, 2026
Watch the whole thing, not just the clips TK mentions one insight from Uber that is just so relevant for every AI founder trying to win the race and accelerate at scale right now -- "You must be able to empower teams. The fewest number of rules while staying out of chaos. https://t.co/xWwWTNi87m
— Ann Bordetsky (@annbordetsky) Mar 15, 2026
There’s a fundamental difference between taking an existing process and applying AI agents to it vs. taking a process from scratch and designing it from the ground up for AI agents. The gap we’re going to see will widen between the teams and companies that are able to do the
— Aaron Levie (@levie) Mar 14, 2026
I've reached a point where AI outputs are generated too rapidly, and they are beyond my own "pay grade"! This is why I have it generate slide decks instead! I feel like that CEO who doesn't quite understand what his technical minions are conjuring up! https://t.co/IBNu1PnpBa
— Carlos E. Perez (@IntuitMachine) Mar 13, 2026
For folks running autoresearch: here are the top 10 findings from 20+ agents across 1000+ experiments. 1. Step count dominated everything 2. A simple attention pattern consistently won 3. Initialization turned out to matter more than optimizer tweaks 4. The swarm discovered a https://t.co/pzOMqCtonM
— Christine Yip (@christinetyip) Mar 13, 2026
Travis Kalanick built a global company in total secrecy for 8 years. Thousands of employees across 30 countries, and none of them were allowed to put the company name on LinkedIn. He just came out of stealth, launched Atoms, and dropped an hour with @tbpn. One of the best https://t.co/y6SEac3Sdg
— Anish Moonka (@AnishA_Moonka) Mar 14, 2026
This looks cool, bookmarking for later https://t.co/ZBvXCN56LG
— Garry Tan (@garrytan) Mar 14, 2026
Introducing autocontext: a recursive self-improving harness designed to help your agents (and future iterations of those agents) succeed on any task. I built this for our clients with the intention of commercializing it but the community support around Karpathy's https://t.co/QSqyvklZrZ
— Jay Scambler (@JayScambler) Mar 13, 2026
Been saying this for a year. Agentic AI is backend engineering far more than it is AI. This stands true for any technology, once you scale and abstract it enough, you’re only left with engineering problems. Learn > Event driven systems > Data pipelines > Distributed systems > https://t.co/grUe29qW9f
— Aarno (@TheGlobalMinima) Mar 14, 2026
FULL INTERVIEW: @travisk joins TBPN to discuss his new company Atoms, physical AI, Uber, and more: 01:18 - Why he's been building in stealth for 8 years 04:32 - Atoms and the future of physical AI 08:10 - Creating a culture of builders 12:05 - Lessons from Uber 24:30 - The https://t.co/ZJd0jbUCwn
— TBPN (@tbpn) Mar 13, 2026
Atoms. https://t.co/nbCnIkI3RD
— travis kalanick (@travisk) Mar 13, 2026
PR with the details https://t.co/OBJTzWz63z
— tobi lutke (@tobi) Mar 12, 2026
Oh nothing, just the CEO of @Shopify using @karpathy Autoresearch technique to improve a templating engine that was in production for 20 years by *checks notes* 51%! Truly... this, applied everywhere is how we foom https://t.co/fhHBYrUbj8
— Alex Volkov (Thursd/AI) (@altryne) Mar 12, 2026
https://t.co/LeaTneJ9f5
— Manthan Gupta (@manthanguptaa) Mar 13, 2026
My CTO friend texted me: "Your gstack is crazy. This is like god mode. Your eng review discovered a subtle cross site scripting attack that I don't even think my team is aware of. I will make a bet that over 90% of new repos from today forward will use gstack." https://t.co/P7aOFu5wFM
— Garry Tan (@garrytan) Mar 12, 2026
A MUST-read interview with an Accenture employee explaining the current usage patterns of $MSFT Copilot, Claude, and other AI models in enterprise usage: 1. They mostly work with large and large mid-market companies. According to him, as far as it comes to $MSFT Copilot, https://t.co/toiA1p8v0g
— Rihard Jarc (@RihardJarc) Mar 13, 2026
Karpathy’s Autoresearch is not another OpenClaw clone. In fact, it’s fundamentally different, with applications that extend far beyond being the “muscle” of a personalized AI assistant. Moreover, its implementation could vary widely across disciplines, potentially advancing https://t.co/VMdOf9JP7M https://t.co/WhNAGuRIuW
— Eli5DeFi (@Eli5defi) Mar 12, 2026
Expectation: the age of the IDE is over Reality: we’re going to need a bigger IDE (imo). It just looks very different because humans now move upwards and program at a higher level - the basic unit of interest is not one file but one agent. It’s still programming. https://t.co/4YD3dzuf4d
— Andrej Karpathy (@karpathy) Mar 11, 2026
New @openclaw beta bits are up! With Hunter🏹 Alpha (1M context!) and Healer🩹 Alpha FREE stealth models from @OpenRouter Also, GPT 5.4 and @Kimi_Moonshot Coding now are more reliable, and lots of fixes around ACP and message handling. https://t.co/Z3roxNBoJK
— Peter Steinberger 🦞 (@steipete) Mar 12, 2026
https://t.co/BeMJmb8k2w
— John Coogan (@johncoogan) Mar 11, 2026
https://t.co/YwlMDBaTr9
— Aravind Srinivas (@AravSrinivas) Mar 11, 2026
I've been using OpenClaw for 2 months. This is everything I wish I knew when I started. If you still haven't used it - this video is for you. "The Ultimate Beginner's Guide to OpenClaw" - link below https://t.co/SuDHUz4Ufp
— Miles Deutscher (@milesdeutscher) Mar 11, 2026
Announcing Personal Computer. Personal Computer is an always on, local merge with Perplexity Computer that works for you 24/7. It's personal, secure, and works across your files, apps, and sessions through a continuously running Mac mini. https://t.co/EpvilVX6XZ
— Perplexity (@perplexity_ai) Mar 11, 2026
this is fucking WILD. Nvidia just dropped an article breaking down how AI works at an industrial level (save this): >AI runs on 5 layers: energy → chips → infrastructure → models → applications >we’re a few hundred billion in. trillions still need to be built >this https://t.co/qviaEBYxge https://t.co/SWZxZvILaM
— Damian Player (@damianplayer) Mar 10, 2026
@JJEnglert @openclaw You’re on Apple you are wrong with the heartbeat. Heartbeat can be done as a launchd command (which is macOS cron job). Take it off the LLM entirely. Do not burn tokens needlessly.
— ArbRoyalsTheAce (@ArbRoyalsTheAce) Mar 11, 2026
After a month of watching my fellow builders set up their @openclaw , I finally took the plunge this past week. Last night my agent ran overnight on a project we came up with together, and it was ready for review when I woke up this morning. It has its own GitHub account. Its https://t.co/M4pDEjNaav
— JJ Englert (@JJEnglert) Mar 10, 2026
If you think @openclaw is all hype, read JJ’s post. It’s insanely powerful if you set it up right. Here’s a quick checklist for implementing his system: □ Install OpenClaw (takes 5 min) □ Create a GitHub account for your agent □ Give it SSH access to push commits □ Set up https://t.co/OUZNL5QjsE
— Corey Ganim (@coreyganim) Mar 11, 2026
@readswithravi https://t.co/kAbeVv3iOG Americana - A 400 Year History of American Capitalism
— Chamath Palihapitiya (@chamath) Mar 10, 2026
AI should help us produce better code https://t.co/q8N9nJyLDk
— Simon Willison (@simonw) Mar 10, 2026
oh yeah i should have linked autoresearch probably https://t.co/MhdH7YF1PP (you don't "use it" directly, it's just a recipe/idea - give it to your agent and apply to what you care about.) and the tweet about it that went mini-viral over the weekend with more context
— Andrej Karpathy (@karpathy) Mar 9, 2026
We're mostly done with 1.0 of our GTM Agent **Salesforce is the CRM **Apollo as contact search/updates **Apollo as the sequence manager **RepVue data as the signal drivers (we are lucky enough to have our own proprietary intent data) **Gmail as the email **Chorus for raw call https://t.co/k0f8tuuzFD
— Ryan @ RepVue (@ryan_c_walsh) Mar 10, 2026
@FDavidsonT has launched over 5,000 agents and used over 3.5B tokens *just for himself* in the last 31 days alone. This is what 10x productivity looks like in practice. https://t.co/2RiZhSVawh https://t.co/mhmRlpoJTH
— Alfred Lin (@Alfred_Lin) Mar 10, 2026
https://t.co/Lb2gtM9SNs
— Viv (@Vtrivedy10) Mar 10, 2026
Autoresearch applies to a lot more areas than ML, below is a short list of 20 applications areas for founders and builders: what other areas you think it applies to? https://t.co/yLOyfkXtEu https://t.co/FZt0o4pyA9
— Misbah Syed (@MisbahSy) Mar 10, 2026
Read about AlphaGo’s incredible impact over the past 10 years and our vision for the future: https://t.co/utLIKwSGqg
— Demis Hassabis (@demishassabis) Mar 10, 2026
What if you could optimize a model overnight without any ML experience? What if an AI agent runs hundreds of training experiments autonomously, keeping only the improvements? That is the idea behind autoresearch. Yes, the early results are small scale, GPT-2 speedups, a 0.8B https://t.co/MbkFEYgBbq
— Philipp Schmid (@_philschmid) Mar 10, 2026
Our GTM agent is easily the single most widely used and loved agent we’ve built internally The GTM agent has: - saved reps >40hrs per month, each - lead-to-qualified-opportunity conversion rate up 250% - 50% of reps use it daily and 86% use weekly Want to work on similar https://t.co/5j0xDJMh2a
— Brace (@BraceSproul) Mar 9, 2026
https://t.co/9WLIjS8Lrk
— LangChain (@LangChain) Mar 9, 2026
🚨 Holy shit.. Karpathy just open-sourced a repo that lets AI agents run their own ML research experiments autonomously. 630 lines of code. Single GPU. No human in the loop. Every dot in his graph is a full LLM training run that took 5 minutes, and the agent just keeps going, https://t.co/Aunm5NKja4 https://t.co/RsPSgaR5uL
— Alex Prompter (@alex_prompter) Mar 9, 2026
What this means for non-developers: Claude Code now automatically reviews code changes before they go live. When someone submits new code, a team of AI agents checks it for bugs. The business application: automated quality control. Imagine this same concept applied to: → https://t.co/VIZjDH4fTn
— Corey Ganim (@coreyganim) Mar 9, 2026
Three days ago I left autoresearch tuning nanochat for ~2 days on depth=12 model. It found ~20 changes that improved the validation loss. I tested these changes yesterday and all of them were additive and transferred to larger (depth=24) models. Stacking up all of these changes, https://t.co/j34dSt4oht
— Andrej Karpathy (@karpathy) Mar 9, 2026
Alpha! https://t.co/QL8Irgp4Hf
— Marc Andreessen 🇺🇸 (@pmarca) Mar 9, 2026
interesting and hiring: @joannejang - gm of labs at openai ("inventing and prototyping new interfaces for how people collaborate with ai") - stanford bs mathematical & computational sciences and ms cs - interned with congressman Adam Schiff, at nasa jpl, at apple, at disney, and https://t.co/jJfFwTE3jy
— Chris Barber (@chrisbarber) Mar 9, 2026
This is gold. If you install Claude on your machine and want to have a proper setup with Cowork because you feel it's the right tool for you, follow these instructions. https://t.co/TIXcnzFavx You will spend a few hours discussing with Claude about you and your business. But
— Nikos Drandakis (@drandakis) Mar 8, 2026
Had my @openclaw agent read the whole thing. Here’s what matters: → Skills work across Claude, Claude Code, Cowork, AND the API → The description field is everything (include WHAT + WHEN) → Use progressive disclosure (don't load everything upfront) → Test for triggering AND https://t.co/7XuIrPbfzi
— Corey Ganim (@coreyganim) Mar 8, 2026
Considering switching from opus-4.6 to gpt-5.4 for your agent? Make sure you read this first https://t.co/2DQu7KBajT
— Ryan Carson (@ryancarson) Mar 7, 2026
Holy shit, this is INSANE. Andrej karpathy just open-sourced an autonomous AI researcher that runs 100 experiments while you sleep > He built a free AI agent that improves itself while you sleep > it runs fully autonomously, executing real experiments with no supervision > https://t.co/sqNZA2HfKf https://t.co/bzmmJFUMal
— CG (@cgtwts) Mar 8, 2026
OK this thing is totally insane. Before going to bed I... * used try to make a new qmdresearcher directory * told my pi to read this github repo and make a version of that for the qmd query-expansion model with the goal of highest quality score and speed. Get training data from https://t.co/hbCfD62ElJ
— tobi lutke (@tobi) Mar 8, 2026
https://t.co/O7OiUYKjbh
— Aaron Levie (@levie) Mar 8, 2026
I packaged up the "autoresearch" project into a new self-contained minimal repo if people would like to play over the weekend. It's basically nanochat LLM training core stripped down to a single-GPU, one file version of ~630 lines of code, then: - the human iterates on the https://t.co/3tyOq2P9c6
— Andrej Karpathy (@karpathy) Mar 7, 2026
Karpathy just open-sourced autoresearch. One GPU. 100 ML experiments. Overnight. You never touch the code — just write a Markdown file. The bottleneck isn't compute. It's your program.md. https://t.co/zcR77tlOgO
— Garry Tan (@garrytan) Mar 8, 2026
Sandboxes are layer one. As agents take on more complex work, every layer needs rethinking: - Networking for agent to agent communication - Storage for petabyte scale snapshots - Observability for debugging million path execution trees - Security for autonomous decision making
— Ivan Burazin (@ivanburazin) Mar 6, 2026
https://t.co/AsFDJlRzlt
— Mustufa Khan (@mustufa4socials) Mar 7, 2026
Brilliant economic paper directly models the "Structural Jevons Paradox" happening right now in the AI industry. The cost of running an LLM is dropping, but total computing energy is exploding anyway. It mathematically proves that as the unit cost of digital intelligence and https://t.co/825YCwxjZ2 https://t.co/795S4jWcjD
— Rohan Paul (@rohanpaul_ai) Mar 7, 2026
An update toward AI systems being able to reliably book flights this year (even if required to use legacy UIs). (Previous discussion: https://t.co/SCdLA9o00T) https://t.co/axJ2CL9HME
— Steve Newman (@snewmanpv) Mar 6, 2026
Under Secretary of War for Research and Engineering Emil Michael on being the CTO of the Department of War, applying lessons from Silicon Valley at the Pentagon, and his "holy cow" moment with AI vendors. 00:00 Silicon Valley to DC 02:03 Why the DoW cannot operate at "peacetime https://t.co/8BnP5yrdzI
— a16z (@a16z) Mar 6, 2026
Awesome job by the @databricks team My summary: They trained a model called KARL that beats Claude 4.6 and GPT 5.2 on enterprise knowledge tasks (searching docs, cross-referencing info, answering questions over internal data), at ~33% lower cost and ~47% lower latency. The key https://t.co/EWwiuverUX
— Jamin Ball (@jaminball) Mar 6, 2026
Boris Cherny (Head of Claude Code, Anthropic) just dropped ~90 mins on Lenny's Podcast about what happens after coding is solved. Just the clearest thinking I've heard on where software is actually going. My notes: 𝟭. 𝗖𝗼𝗱𝗶𝗻𝗴 𝗶𝘀 𝗹𝗮𝗿𝗴𝗲𝗹𝘆 𝘀𝗼𝗹𝘃𝗲𝗱. Boris has https://t.co/pw5om66Kex
— Anish Moonka (@AnishA_Moonka) Mar 6, 2026
Google just dropped another banger 🤯 They open-sourced the Agent Development Kit, and it perfectly pairs with Gemini 3.1 Flash-Lite. Means you can now build always-on AI Agents that run 24/7 at a negligible cost. 100% Open Source. https://t.co/1yUISHFkZz
— Simplifying AI (@simplifyinAI) Mar 6, 2026
I took my normal "exit plan mode" skill and expanded it into a "mega exit plan mode" skill that I find can be quite useful when you are trying to expand scope and/or just want Claude Code to delight you with new ideas. It's sort of like the expansive CEO to the normal skill's
— Garry Tan (@garrytan) Mar 6, 2026
Don't overload your AGENTS dot md files. Keep them brief. GitHub's analysis of 2,500+ repos found what makes AGENTS dot md files work: - provide the agent a specific job or persona - exact commands to run - well-defined boundaries to follow - clear examples of good outputs https://t.co/GFSEjLsx8C
— elvis (@omarsar0) Mar 5, 2026
Anthropic's Revealing Chart on AI's Impact on Jobs Anthropic has unveiled a pivotal chart that underscores the chasm between AI's capabilities and its real-world application in the workforce. Derived from analyzing 2 million actual conversations with Claude, this radar chart, https://t.co/eu8E7D1zlL
— Brian Roemmele (@BrianRoemmele) Mar 6, 2026
Google finally did it □ Stop scanning 47 emails to find the 3 that matter - your agent triages your inbox before you wake up □ Walk into every meeting prepared without doing the prep - agenda, attendees, and docs pulled automatically □ Plan your entire week in 90 seconds - https://t.co/svdmLYWU2q
— Nick Spisak (@NickSpisak_) Mar 5, 2026
“If a task is already outsourced, it tells you three things. One, the company has accepted that this work can be done externally. Two, there’s an existing budget line that can be substituted cleanly. Three, the buyer is already purchasing an outcome. Replacing an outsourcing https://t.co/fmB74AV7t5
— Aaron Levie (@levie) Mar 6, 2026
https://t.co/hsKrbn1gEq
— Julien Bek (@JulienBek) Mar 5, 2026
The Brand Age: https://t.co/U6z3g6ZQGd
— Paul Graham (@paulg) Mar 5, 2026
My deep dive article on $AMZN and why it is my biggest position: 1. Value of e-commerce in agentic shopping 2. Disruption of the Ad business 3. Agentic AI lead: MCP server adoption showing a strong lead for AWS vs peers 4. AWS (growth of traditional and AI compute). 5.
— Rihard Jarc (@RihardJarc) Mar 5, 2026
Still true two years later https://t.co/43sl4a7haw
— Vinod Khosla (@vkhosla) Mar 4, 2026
I’ve never seen an agent orchestration system that operates across all business functions And, neither have I seen one that has been built with great taste i.e. design and UX quality like Linear etc Paperclip seems to be special - no agent tie in either https://t.co/BlFcrLLqtF
— Numman Ali (@nummanali) Mar 4, 2026
NotebookLM just wiped out a whole bunch of video designers with a casual launch. https://t.co/EKpTAklH7U https://t.co/z5DfLGBPoj
— CG (@cgtwts) Mar 4, 2026
https://t.co/uSKSJZhZIc
— swarit (@swaritjoshipura) Mar 4, 2026
Banger article for anyone wanting to turn Claude into an actual employee! Covers: 1/ What Cowork actually is (and isn't) 2/ How it's different from Chat and Claude Code 3/ Full setup in 15 minutes (folder structure + context files) 4/ Which plugins to install first 5/ 3 https://t.co/JNT267QW3z
— Corey Ganim (@coreyganim) Mar 3, 2026
🚨 Someone just solved the biggest bottleneck in AI agents. And it's a 12MB binary. It's called Pinchtab. It gives any AI agent full browser control through a plain HTTP API. Not locked to a framework. Not tied to an SDK. Any agent, any language, even curl. No config. No https://t.co/naBiwu0sAe
— Nav Toor (@heynavtoor) Mar 3, 2026
New @openclaw beta is up and it's MASSIVE. https://t.co/jL9ZNIefRc Ask your agent to update to beta or run: curl -fsSL https://t.co/yjjVIN9fzw | bash -s -- --beta
— Peter Steinberger 🦞 (@steipete) Mar 3, 2026
https://t.co/g9OEwZxKYk
— Pontus Abrahamsson — oss/acc (@pontusab) Mar 3, 2026
We have only seen the tip of the iceberg for layoffs... "A company like Monday, let's say we have like $1BN in expenses. Probably spend, 60%-70% of the budget on headcount and then 7%-8% on software. If we flip the equation and scale the business not on headcount but with https://t.co/gXUDaAllP2 https://t.co/etOI92743J
— Harry Stebbings (@HarryStebbings) Mar 2, 2026
"Not a single one of our full-time engineers right now writes code anymore... it's all agentic... the way you build a company completely changes." @justindross with @JTLonsdale on the future of software: JD Ross: Not a single one of our full-time engineers right now writes https://t.co/o97Td9q8IT https://t.co/TJ8PiIwKV2
— American Optimist (@AmOptimistShow) Mar 2, 2026
Brilliant post by CEO of Cursor. Your next software engineering teammate isn't human. - The first era of AI software development relied on Tab autocomplete to automate repetitive, single-keystroke coding tasks. - The second era introduced synchronous agents guided by https://t.co/gsjOOhO7WA https://t.co/QUxcNe2TXy
— Rohan Paul (@rohanpaul_ai) Mar 2, 2026
The force multiplier of the agent harness right now is crazy. The industry has landed on some architectural consistency, but there are still so many different variants of how to attack this. Maybe this gets bitter lessened out of existence, but for now it’s a huge lever. https://t.co/J05na51GA9
— Aaron Levie (@levie) Mar 3, 2026
This paper buries the most important claim in AI infrastructure in a single phrase on page 8: “LLM-as-Operating-System paradigm.” Read that again. The authors are arguing that LLMs are becoming the next operating system layer, and context engineering is the systems programming https://t.co/cit5yPi1E4
— Aakash Gupta (@aakashgupta) Mar 2, 2026
> Anthropic just launched a free AI academy with 13 certified courses across MCP, APIs, Claude Code and more. > it’s completely free https://t.co/UvhURsxbTI https://t.co/d9aJHYVYmc
— CG (@cgtwts) Mar 1, 2026
@stratechery @benthompson https://t.co/pJ45ksZejQ
— Carlos E. Perez (@IntuitMachine) Mar 2, 2026
Anthropic and Alignment Anthropic is in a standoff with the Department of War; while the company's concerns are legitimate, it position is intolerable and misaligned with reality. https://t.co/RK9q4URnTA
— Stratechery (@stratechery) Mar 2, 2026
This article is just soooo niche man… check it out if you are thinking of training tool-use or terminal-use agents yourself. https://t.co/NKsxoVgw70
— AVB (@neural_avb) Mar 1, 2026
🚨McKinsey just dropped how to build agentic AI (that works) Here's everything you need to know in 2 minutes: https://t.co/hDBUkZmM20
— Matt Dancho (Business Science) (@mdancho84) Mar 1, 2026
Men only want one thing and it's to embody the spiritual state of an old Japanese man making pizzas while chain smoking cigarettes in a small restaurant in the forests of Japan. https://t.co/gOSpqg7tT4
— Autism Capital 🧩 (@AutismCapital) Mar 2, 2026
Is Traditional Software Engineering Dead? “Does this mean that traditional software engineering is dead? Absolutely not. Software engineers—even the ones who are not necessarily tuning or training AI models—these are now among the most leveraged people on earth. Sure, the guys https://t.co/Aw8Uk8Bj0b
— Naval (@naval) Mar 2, 2026
Great article, and yes, confirmed, the MCP hype is over. https://t.co/VlQFzmJwsw
— Morgan (@morganlinton) Mar 2, 2026
https://t.co/CYqPTytV7g
— witcheer ☯︎ (@witcheer) Feb 28, 2026
🚨 BREAKING: Alibaba just handed the AI agent community a production-grade sandbox for free. OpenSandbox is a full-stack platform for running untrusted agent code safely: → Unified APIs across multi-language SDKs → Docker and Kubernetes runtimes purpose-built for agents → https://t.co/kEmZ2Po629
— Muhammad Ayan (@socialwithaayan) Feb 28, 2026
