AI is quietly becoming one of the strongest edges in crypto. Most people still research the old way — scrolling charts, reading endless threads, and reacting late. I’ve been building simple AI workflows that cut research time and help filter noise earlier. Here’s the 3-step approach I actually use ↓

Our design language has evolved since the last reveal: from S3XY to Cybertruck, Cybercab & Optimus Roadster is inspired by the SR-71 & F-35

When you schedule invites to an X call, you can send participants an optional invite to their email. It will come from notify@x.com . X should probably include the username, it looks like X is inviting people when I send them, lol. check it out at call.x.com !

🚨NEW EXPERIMENT 🚨 Playground is an experimental gaming platform that lets you create your own games with zero coding experience. If you can think it, you can play it. Go to labs.google/playground to learn more! Available to users 18+ in the US.

🇺🇸 I’m not sure there has ever been a table quite like the one Trump assembled at the White House today. Elon on his left. Jensen Huang on his right. Zuckerberg, Pichai, Bezos and many of the people actually building the AI economy around them. And Trump’s message to the room was remarkably simple: “We have a very big lead, and we’re gonna keep our lead.” Washington now explicitly talks about AI as national power. Trump says it could become bigger than the Industrial Revolution, and he increasingly describes the competition with China as one with a winner and no meaningful second place. A week after Xi sat down with many of these same technology leaders, they’re back in Washington together. I think we’re watching Silicon Valley become part of American statecraft in real time. Writer: Oliver x.com/EricLDaugh/status/2105…

Manus launched a pile of new products and updates under the Manus 2.0 umbrella. > Cascade & new foundation: Manus 2.0 introduces a more efficient agent architecture with Cloud Computers and event-triggered Automations. > Manus Studio: Manus desktop app becomes a professional AI workspace with new Video Editor, Alchemy, Game Dev, Remote Control, and Computer Use capabilities. > Cue: A new standalone app for personal agents with their own email, phone number, wallet, computer, and multi-agent collaboration. I need testing time 👀

We are building the infrastructure of the future, from supercomputers to Terafab to Gigasat to a new Starbase in Louisiana

BREAKING: Starship just deployed the first-ever Starlink V3 satellite into operational orbit. History made. This is the beginning of a massive new era for Starlink. 🚀

Introducing Claude Opus 5.5, the first model in our new Claude 5.5 family. It performs at the level of Claude Fable 5.1 for most tasks, and costs 40% less to run than Opus 5.

Grok Bot can talk now.

Grok Build now gets better the more you use it. It remembers conventions, decisions, and project facts across sessions. /memory to browse /dream to organize recent notes into topics x.ai/build

I’ve just gone through Anthropic’s September threat intelligence report and there’s something in there I think people are underestimating. We already know hackers are using AI. That’s not really news anymore. What caught my attention is how little the human sometimes needs to do now. Anthropic documented operations where AI was being used across reconnaissance, exploitation, credential harvesting, processing stolen data and even modifying tools after security systems detected them. In one of the financially motivated operations they investigated, Anthropic says AI agents did nearly all the work. Think about that for a minute. We’ve spent the last few years thinking about AI mainly as something you ask questions or use to help you do a task. But we’re moving towards something different: you give the AI an objective, it figures out the steps, uses different tools, executes the work and comes back when it needs you. Obviously Anthropic’s report is showing the dangerous side of that. But while reading it, I kept thinking about the other side. If an agent can coordinate a complicated cyber operation, what happens when you give the same level of autonomy to normal business operations? Instead of opening your accounting software and checking 10 different reports, imagine telling it: “Close August and show me anything that needs my approval.” Or: “Go through our overdue invoices, figure out who needs following up, prepare everything and bring me the important ones.” That, to me, is where business software is heading. Not adding a chatbot to every SaaS product and calling it AI. The software itself starts doing the work. There’s a catch though. The more access we give these agents — email, bank transactions, customer records, accounting, files, payments — the more dangerous bad permissions become. So I don’t think the race is simply going to be about who builds the most autonomous agent. It’ll also be about who can make that autonomy safe: proper permissions, limits, audit trails and human approval where it actually matters. This Anthropic report is about cybersecurity, but I think it accidentally gives us a pretty good preview of where software in general is going. We’re moving from software we operate to software that operates for us. That’s a much bigger change than just putting AI inside existing apps.

OpenAI has launched ChatGPT for Financial Services—and this is much bigger than simply giving Wall Street a smarter chatbot. The product is designed for investment banking and equity research. It combines GPT‑6 Astra with specialist financial datasets from companies such as LSEG, PitchBook, Daloopa, Crunchbase and Quartr. It can help analysts conduct research, build financial models and prepare pitchbooks, while offering enterprise controls such as role-based access, encryption and audit-log exports. Morgan Stanley and Evercore reportedly helped OpenAI design it around real financial workflows. Why does this matter? Because the most valuable AI products may not win by having the smartest model alone. They will win by combining five things: A capable AI model Trusted industry data Real professional workflows Security and accountability Distribution inside established businesses The underlying model can eventually become cheaper or easier to replace. But the data relationships, workflow integration, permissions, audit history and customer trust are much harder to reproduce. This also changes the conversation about jobs. It doesn’t necessarily mean investment bankers and research analysts disappear tomorrow. However, the repetitive work traditionally assigned to junior analysts—collecting information, comparing companies, updating models and assembling presentations—could be completed much faster. The human role then shifts toward verifying the AI’s work, exercising judgment, managing client relationships and accepting responsibility for the final decision. It also puts OpenAI closer to companies such as Bloomberg, financial-data providers and specialist enterprise-software businesses. The competition is no longer only about who has the best chatbot. It is about who controls the data and workflows professionals depend on every day. The broader lesson for entrepreneurs is clear: Don’t just place an AI chatbot on top of an existing business. Find a valuable industry workflow, connect the right data, build clear permissions and audit trails, and solve the complete job. The biggest AI businesses may not own the best model. They may own everything surrounding it.

I resigned from Anthropic today. I spent the last three years doing pretraining research at both OpenAI and Anthropic. Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. More thoughts below.

Everything you need to master in 2026 to get rich: • Using Astra 6 high for planning, low for execution • Grok Bot to manage day to day • Coding loops • Taking advantage of Codex's computer use • Herdr to manage tons of agents • /goal • Running local models • Karpathy's Autoresearch • Building an X audience • Creating videos • How machine learning works • How databases work (been using Convex) • Using API's • Training your own LoRAs • Elimination of doom scrolling • Most importantly: focus

Omarchy is the only operating system in 2026 built for AI If you take AI serious, you're making major mistake by not pulling a beater laptop out of your closet and installing Omarchy ASAP Here are 11 tips to get started: 1. Set up is painfully easy. Plug a USB thumb drive (I bought one for 10 bucks at Micro Center) into your normal computer. Go to any AI agent and say to flash the USB drive with Omarchy so you can install it on another computer. Tell it to walk you through step by step how to do it (it's dead simple and takes under 5 minutes) 2. You can use ANY computer. You don't need to buy a new one. There are people installing Omarchy on 15 year old laptops with 2gb of memory. Go to your closet and pull out your old college laptop. 3. Anything you don't like about the OS, just change it. Because the operating system is open source, you can edit the code. Anything you don't like (I didn't love the notification system) I just go to my chatgpt, hermes, or grok bot agents and just say "change this for me". Boom like magic it changes 4. Explore the plugins marketplace. There are some INCREDIBLE community made plugins for the OS. My favorite is a google calendar one that always shows my next meeting in my nav bar. 5. Build your own plugins. This is my favorite part about using open source software. It's never been easier to build your own plugins and contribute to the community. I'm building my own focus timer I'll ship in the next couple of days. 6. Make it your own. The built in themes to Omarchy are BEAUTIFUL. Explore them and see which one fits your vibe the most. I love Osaka nights. 7. Daily drive it. Once you get used to the keybinds and the way the OS works I promise it'll be impossible to go back to MacOS or especially Windows. Give yourself some time to learn the ins and outs. 8. Contribute to the community. Share your tips and tricks on social media. The more people we have using Omarchy, the more people build improvements, the better the OS is. 9. Customize the keybinds. Keybinds are the most important concept of Omarchy. It's an OS built to be used just with your keyboard. So make the keybinds your own. For instance I made super+A a pop up that allows me to quickly add a todo item to my list. Boosts my productivity a ton. To customize, just ask your AI agent to change them 10. If you are going to buy a new computer, I LOVE the Framework 13 Pro, altho the Dell XPS 14 is great too. I like the Framework a lot because it's MacBook Pro quality, but you literally have to build it yourself which made the computer really feel like my own (I'm not sponsored by anyone involved with this post btw) 11. Contribute to the community. The more people using Omarchy the better.

This is GPT-6 Astra. Anything you can do on a computer, Astra can do for you. Fast.

Exchanges selling “safe AI agents” because the bot cannot withdraw are selling the wrong promise. No-withdrawal stops an exit. It does not stop a loss. I separate agent guardrails into three layers: 1. Custody Can it move assets off the venue. Hard deny is the floor, not the product. 2. Execution What it may buy, sell, leverage, and convert. Written scopes. Not vibes in a system prompt. 3. PnL What still holds when the model is wrong. Size, hours, max drawdown. Outside the model. If a launch leads with layer 1 and stays silent on 2 and 3, I treat it as marketing. I do not treat it as risk design.

Grok Bot can now buy things on the internet. To get started, connect @link, send your Bot shopping, and it can complete purchases on your behalf.

NVIDIA just made the agent stack more explicit. SpaceXAI is deploying NVIDIA Vera — billed as the first CPU built for agents — to run the work around the model: orchestration, code execution, data processing, tool use, simulation. The useful distinction: GPUs generate. Agents also need a fast control plane between calls. Inference is only one part of an agent loop. The rest is scheduling tools, running code, moving data, and keeping the GPU fed. Vera is aimed at that second layer. Why that matters: 1. Faster agent loops NVIDIA claims up to ~1.8x faster task completion vs typical x86 on agentic, RL, and data-processing workloads. Sandbox-style work is one of the bigger gains. 2. Higher memory bandwidth Up to ~1.2TB/s LPDDR5X bandwidth. That helps when the bottleneck is moving context, tool outputs, and intermediate state — not just matrix math. 3. Better GPU utilization If the CPU side of the loop is slow, GPUs sit idle between steps. A faster orchestration path keeps more of the expensive silicon doing useful work per watt. 4. Purpose-built cores 88 NVIDIA Olympus cores with Spatial Multithreading. Designed for the concurrent control work agents actually do, not just general server traffic. 5. Cleaner system design Separate the generator (GPU) from the controller (CPU). That makes agent runtimes easier to scale, profile, and constrain. SpaceXAI is also expanding Grok infra on Vera Rubin and planning to take an optimized Vera Rubin NVL72-class system into orbit with its first Starmind satellite. So the story is not “another chip announcement.” It is the industry admitting that agentic systems bottleneck on more than model flops. Capability still needs a place to run the loop. Authority and failure modes still decide whether that loop is safe.

Agent trading products fail the same test. Can you state, in one sentence, what happens when the agent is wrong? If the answer is vague, you do not have a system. You have a demo with a wallet attached. A real setup answers four things before the first live order: 1. Where does the money sit Main account or capped subaccount. 2. What can it do alone Read, propose, or execute. 3. What is the hard ceiling Size, venue, frequency, counterparties. 4. How do you stop it One revoke path, not a support ticket. If those are missing, “AI trading” is just leverage with extra steps.