Yuma

Built this digital twin feature. Created my digital twin in 30s. I then let it (him?) create a personal website in under two minutes. Now my digital twin creates websites, does analyses, create slides and images, promotes my business, number crunch, fill forms, automate ops. Now I only have to tell my twin agent that I don't like bright green.

@fordhsmith Always

Anything that grows a little too quick without the fundamentals Medvi Delve Cluely Some weird agent company startup with terrible UI and useless agents People completely lost sight of the fundamentals in favor of getting a quick hit on socials

49.6% of internet traffic is now bots. 30-40% of your Insta feed is from unfollowed accounts. AI does the news. Algorithms changed and people didn't notice. We wrote a 10,000-word deep dive on how every major platform's algorithm actually works in 2026. Some interesting findings: > X's algorithm scores tweets using 48,000 features. Likes = 30x weight. Retweets = 20x. But the real shift is Grok now WRITES trending topic summaries that hundreds of millions read. X went from distributing human speech to publishing AI-generated editorial content sourced from human speech. > LinkedIn killed "broetry." Creators who built audiences on viral hooks saw 50-70% reach declines. The algorithm now uses NLP to classify whether your post contains genuine expertise. It literally reads your post and decides if you know what you're talking about. > TikTok captures 55-60 minutes per day per user. YouTube: 45-50. LinkedIn: 7 minutes. Total social media time has plateaued at ~2h 23min globally. The algorithms got better. Humans didn't get more hours. > Meanwhile Google AI Overviews reduce click-through to websites by 30-70%. > Stack Overflow lost 14% of traffic. Chegg stock collapsed 99%. The pattern: Every platform is converging on the same architecture. AI generates content, AI ranks content, AI summarizes content, AI consumes content. Humans went from being the audience to being the training data. Full report: o-mega.ai/articles/state…

Anthropic accidentally shipped 512,000 lines of Claude Code's source code via npm. A missing .npmignore line. 1,900 TypeScript files. 3.1 million views on X within hours. We went through the actual codebase, assisted by Claude himself: The core engine that powers every interaction is ~50 lines of code. Think. Act. Observe. Repeat. That's it. No workflow graphs. No state machines. No complex orchestration. The model decides everything. The harness just provides tools and guardrails. Meanwhile, the Bash security system alone is 2,592 lines. 30+ regex patterns, tree-sitter parsing, Zsh module attack prevention, AND a secondary LLM call to validate commands. 50 lines to run intelligence. 2,592 lines to keep it from destroying your machine. That ratio tells you everything about where we actually are with AI agents. 90% of the codebase was written by Claude Code itself. 60-100 internal builds per day. One file has a 5,594-line function with 486 branch points. And hidden behind 32 feature flags: an always-on ambient assistant, a 30-minute deep reasoning mode, a memory consolidation system inspired by sleep research, and a full Tamagotchi pet with 18 species and deterministic gacha mechanics. The biggest number: 50 lines of core intelligence wiring. 512,000 lines of product engineering. Building an AI agent is 1% model integration and 99% making it safe, fast, and useful.

Mistral AI raised $830 million this week. Not from VC. But from 7 banks. Buying 13,800 GPUs for a French data center. This is the story of zero dilution EU AI: Europe's late-stage AI funding gap with the US is 9:1. 🇪🇺 European VCs put 12% of capital into AI 🇺🇸 American VCs put in 76% Europe was and still is losing at venture money. So Europe stopped trying to win the venture game and started playing a different one. JPMorgan projects $300 billion in AI data center deals per year. Morgan Stanley's debt revenue jumped 93% last quarter. Banks are tripping over each other to lend against GPU clusters. Because a data center with long-term enterprise contracts is not a startup. It's a toll road. > Telco investor American Tower carries $40 billion in debt against cell towers. > AI computing company CoreWeave borrowed $7.5 billion against GPUs. Same asset logic. Different decade. Cell towers were once "exotic collateral." Then aircraft. Then fiber. Then data centers. Now GPUs. Every generation of critical infrastructure starts as too speculative to lend against and ends as a standard asset class. Europe didn't need to out-raise Silicon Valley. It needed to recognize that the thing Silicon Valley treats as a startup is actually a power plant. And Europe has been financing power plants for a very long time: o-mega.ai/articles/eu-ai…

@atmoio 😭 ‘Who’s a good boy, who’s the pinnacle of consciousnesses’

Normal people are starting to go crazy

Tried this and it’s pretty crazy, it can generate text based on your questions

United States of Europe

The Netherlands is the third biggest tech capital per capita country in the world Taiwan, US, Netherlands should merge or something

@adamghaida Model incoming

jesus what is happening

@idia by building something in europe

If you're building in Europe, you're the chronic underdog. Not raising 10 million dollar rounds. No rich friends who can be your angels. No network in the tech ecosystem who will help your post go viral. You go two years without any traction and you still keep going. It can get really lonely, which makes it even harder. It will be at some point in your life your biggest advantage, but you just don't know it yet. Things are slow at first, and then everything comes all at once. And then the stuff which seemed huge at first seems to be small today. There will be a moment that things will be in your favor. You just have to work a lot harder for it than others. It doesn't always seem to make sense, but you're probably not in this because it all makes sense. You learn fundamentally different things when you have to do it all from scratch. Keep building Europe 🇪🇺💙🇪🇺

@compliantvc So easy, so true, let’s change that

European founder after raising €12,500 in venture capital

@adisingh You know who

Who’s building OpenClaw for enterprises?? Been following this space closely. Think there’s a big winner that will be born here.

So there’s this AI thing

Robots in Europe, students building what some of the most impressive entrepreneurs in the world are building

Anthropic gave Claude full control of your Mac yesterday. Today you can assign it tasks from your phone and walk away. 4 features in 7 days: > Computer Use (March 23): Claude clicks, types, scrolls, opens apps on your actual desktop. > Projects (March 20): persistent workspaces so Cowork remembers previous tasks. > Code Channels (March 20): control Claude Code from Telegram and Discord. Dispatch (March 17): assign Cowork tasks from your phone via QR code. Pretty lucrative as well: 🤑 Claude Code is at $2.5 billion ARR. Doubled since January 1. 🤑 Anthropic overall: $19 billion revenue run rate. Enterprise AI spending share went from 4% to 40% in one year. They are building an AI operating system on top of your personal machine. Code for developers. Cowork for knowledge workers. Dispatch for mobile. MCP for tool connections. Computer Use for everything the connectors do not cover yet. Here's what most people are missing about how this works. All of it runs LOCALLY ❗ > Your laptop is the server. If it sleeps, Claude sleeps. > Claude operates as you. Your browser. Your email. Your file permissions. If it sends a message, it comes from your OWN account. If it deletes a file, it uses your access. A VC founder asked Cowork to organize his wife's desktop last month. It deleted 15,000 family photos. Recoverable from iCloud, but the point stands: the agent carries your full permissions with no identity boundary. This is a deliberate design choice with real trade-offs. Privacy: your files never leave your machine. That matters. Cost: $20/month for Pro. Hard to beat. Speed: no network round-trips, everything local. But it also means no 24/7 operation unless you keep a dedicated machine running. No separate agent identity. No audit trail separating your actions from Claude's. No way to deploy 10 agents in parallel across cloud infrastructure. The industry is splitting into two models right now: 1. AI assistants that extend you. Claude, ChatGPT. Run on your machine, act as you. Best for individual productivity. 2. AI workforces where each agent has its own browser, its own accounts, its own cloud environment. Runs independently. Best for organizations scaling operations. Both are valid. Both solve different problems. Understanding which one you actually need saves you months of going in the wrong direction. What Anthropic shipped this week tells me we are moving from "AI chat window" to "AI operating system" faster than anyone expected. -- Thoughts as always: o-mega.ai/articles/claud…

@nathanbenaich @airstreet legend!

News! @airstreet has raised $232,323,232 for Fund III to back AI-first companies from the earliest stages in the US and Europe. Now the largest solo GP venture firm in Europe. Our third epoch begins today. Join us!

The Netherlands alone: