News

AI and technology news and observations.

Second attack on Sam Altman's home in days

Early Sunday morning, a shooting occurred near the OpenAI CEO's home in San Francisco — two suspects detained. This is the second attack in just days.

Behind the attacks on Altman's home and the New Yorker profile

A Molotov cocktail, a shooting at his home, and a Ronan Farrow profile in the New Yorker based on 100+ interviews. Altman acknowledged the accusations and called the AGI race a "Shakespearean drama."

Gemini now lets you build interactive visualizations in chat

Gemini's generative interface now renders interactive elements directly in chat. The author used it to visualize a floor lamp's dimensions before buying.

Harvard researchers: AI nearly doubles business revenue

A field experiment by INSEAD and Harvard Business School with 515 startups found that companies which learned to map AI to specific business processes earned 1.9x more revenue than the control group.

Workers in Indian factories wear head cameras to train robots

Workers at Indian factories wear head-mounted cameras whose movement footage is used directly to train humanoid robots.

Unitree humanoid robot reaches 10 m/s — nearly Olympic sprinter speed

Chinese robotics company Unitree demonstrated a humanoid robot running at 36 km/h — for reference, Usain Bolt peaked at 44 km/h.

Digest: April 8–12, 2026

Meta returned to the frontier model race with Muse Spark, Anthropic revealed a model it won't release publicly, OpenAI introduced a mid-tier pricing plan, and CoreWeave disclosed an $87.8B backlog.

Gemini and NotebookLM are now one team

Google has merged Gemini and NotebookLM: notebooks can now be created directly in Gemini, with sources automatically syncing between the two services.

Molotov cocktail thrown at Sam Altman's home

A suspect threw an incendiary device at the OpenAI CEO's home, then made threats at company headquarters. The incident is part of a growing wave of violence against AI infrastructure.

Artemis 2 mission complete: perfect bullseye splashdown

April 11, 2026. Orion splashed down right on target. For the first time in 54 years, four humans returned from the Moon safe and well.

Qwen Chat rebranded as Qwen Studio

Alibaba has officially rebranded Qwen Chat to Qwen Studio, expanding the platform into a developer-focused environment with multimodal capabilities, media generation, and third-party function calling.

Mark Zuckerberg unveils Meta Superintelligence Labs' first model

Meta launched Muse Spark, a multimodal flagship model built from scratch. It already powers Meta AI and will roll out to WhatsApp, Instagram, Facebook, and smart glasses within weeks.

Revolut launches AI agent for personal finance

Revolut has launched AIR, a built-in AI assistant that manages accounts, tracks spending and investments, blocks cards, and buys eSIMs directly in chat. The rollout started in the UK for 13 million customers.

Perplexity launches Portfolio financial AI assistant

Perplexity launched Portfolio, a personal finance agent inside Perplexity Computer that connects bank accounts, brokerages, and credit cards via Plaid to deliver spending analysis and investment portfolio insights.

Digest: April 5–8, 2026

Anthropic introduced an AI model for finding zero-day vulnerabilities and surpassed OpenAI in revenue, Intel began making chips for Tesla and SpaceX, and the Generalist robot showed 99% reliability in physical tasks.

Zhipu AI leads in agentic coding — and raises API prices

Zhipu AI's open-source GLM-5.1 outperformed GPT-5.4 and Claude Opus 4.6 on SWE-Bench Pro, while Z.ai raised API prices by at least 8%, with Alibaba and Tencent following suit.

Media lost half their search traffic to AI while chatbots deliver less than 1%

Chartbeat data shows small publishers lost up to 60% of search traffic over two years, while ChatGPT, Gemini, and Claude together account for less than 1% of pageviews. Meanwhile, Google is testing AI-powered rewrites of publisher headlines in search results.

Telegram lets bots talk to each other

Telegram has lifted its long-standing ban on bot-to-bot communication, enabling bots to interact in groups and via Business Mode. This makes it possible to build multi-bot pipelines inside a single chat without an external orchestrator.

Cognitive capitulation: 73% of people accept faulty AI answers without checking

A University of Pennsylvania study with 1,372 participants and 9,500+ trials found that people massively disengage critical thinking when working with AI — even when answers are obviously wrong.

Claude Mythos: Anthropic released its most powerful model, but not for everyone

Anthropic launched Claude Mythos under Project Glasswing — a closed initiative with AWS, Google, Microsoft and others focused on zero-day vulnerability research. The model scored 93.9% on SWE-bench Verified and autonomously found thousands of vulnerabilities across major operating systems and browsers.

Digest: April 2–5, 2026

Google released the Gemma 4 open model family, Cursor launched version 3 of its agentic IDE, SpaceX filed for IPO at up to $1.7T valuation, and Anthropic acquired a biotech startup for $400M.

OpenAI proposes restructuring the economy for the AI age

OpenAI released 'Industrial Policy for the Intelligence Age,' proposing a public wealth fund, a shift from taxing labor to taxing capital, a 32-hour workweek, and automatic social support triggered by AI-driven unemployment metrics.

A guy built a $1.8B AI company with two employees. Nobody read the fine print

Matthew Gallagher built a $401M telehealth funnel with two employees and NYT crowned him a symbol of the AI future. Behind the headline: fake doctor profiles on Facebook, an FDA warning, and a 1.6M-patient data breach.

Nutella on camera during the Artemis-2 stream: an accidental ad from lunar orbit

A Nutella jar accidentally appeared on camera during the Artemis-2 livestream — Twitter dubbed it the most expensive ad of the century.

I built a special page about the Artemis II Moon mission

A compact mission explainer: who's on board, where Orion is right now, and why the hardest part isn't getting there — it's coming back. All on one page.

Why middle management is obsolete — and how to delegate it to AI

Jack Dorsey published a manifesto on restructuring Block: the company is flattening its hierarchy, replacing middle managers as information relays with an AI system that holds live context across the entire organization.

Anthropic is building an always-on Conway agent

Conway, discovered in claude.ai's code, is a background AI agent that runs without user interaction and accepts requests via webhooks from external services. The Claude Code source leak also revealed KAIROS and ULTRAPLAN.

AI chatbots start renewing psychiatric prescriptions without a doctor in the US

Legion Health, a YC startup, became the first company authorized to use AI for psychiatric drug renewals — the bot extends antidepressant prescriptions for $20/month under Utah's regulatory sandbox.

Digest: March 29 – April 1, 2026

Anthropic accidentally leaked the Claude Code source code via an npm package, OpenAI closed a record $122B funding round, and Google released the budget video generation model Veo 3.1 Lite.

NASA launches the Artemis II moon mission

NASA launched Artemis II — the crew will orbit the Moon over 10 days on a free-return trajectory, covering ~700,000 miles. It's the farthest humans have ever traveled into space.

China launches first humanoid robot assembly line — one robot every 30 minutes

China launched the world's first automated humanoid robot assembly line in Foshan, producing one robot every 30 minutes. In 2025, the world made roughly 16,000 humanoids — 90% of them in China.

Google releases Gemma 4

Google released Gemma 4, an open-source model in four sizes (2B–31B) with up to 256K context and AI agent support. Runs on consumer hardware without a data center.

15% of Americans would accept an AI boss, Quinnipiac poll finds

A Quinnipiac poll of 1,400 Americans finds only 15% would accept an AI supervisor; AI literacy and optimism are moving in opposite directions — the more people know, the less they trust it.

Testing Veo 3.1 Lite: 10 clips for $3, each generated in under a minute

A hands-on test of Google's Veo 3.1 Lite via API: ten eight-second clips cost around $3 total, with each video generating in 40–50 seconds.

Google releases Veo 3.1 Lite — a cheap video generation model

One second of video costs just 5 cents. Text-to-video and image-to-video, 720p/1080p, 4–8 second clips.

OpenAI closes $122B round at $852B valuation

OpenAI completed the largest funding round in history. $852B valuation — less than SpaceX's $1.25T, but no rockets to launch.

Telegram tests built-in AI text editor

Telegram's iOS beta has an AI editor with translation, style adjustment, and error correction right in the input field. Powered by Cocoon on TON.

How to export and transfer chats between AI assistants

A guide to exporting and transferring chats between ChatGPT, Claude, and Gemini.

First time booking flights with AI

An MCP server for Google Flights: connect to Claude, type your route — it searches, filters, and sorts by price. Works via reverse-engineered API.

Digest: March 25–29, 2026

Anthropic considers $60B+ IPO, OpenAI shuts down Sora, Google releases Gemini 3.1 Flash Live, Cursor introduces real-time RL.

Google added chat import from ChatGPT, Claude and others to Gemini

You can upload your full chat archive or ask your old chatbot for a preferences summary and feed it to Gemini.

Goldman Sachs: How to become an AI-native business

K-shaped economy in AI: some companies soar, others sink. Speed of decision-making matters most.

OpenAI and Anthropic preparing next-gen frontier models

OpenAI shut down Sora for model Spud. Anthropic leaked Claude Mythos — a new class above Opus.

We overthink ideas for vibecoded products

An app makes your MacBook moan when you slap it. That's the entire app. The guy made $5,000 in three days.

Why Claude Code tokens run out so fast

A bug found: auto-memory duplicates the entire context with a background call. Plus tips for saving limits.

Chinese AI models surpass American ones in usage

Per OpenRouter, DeepSeek and MiniMax consume more tokens at prices 6x lower than Claude.

Digest: March 22-25, 2026

OpenAI shut down Sora and raised its round to $120B, Anthropic released Auto Mode and Computer Use, Apple revealed Siri relaunch plans, Arm released its own chip.

Tech job market turns up for the first time after three years of decline

Tech companies have 67,665 open engineering positions — up 78% from the 2023 low. Data Science & ML +90%, product and marketing +27%.

Claude Code gets a real Auto Mode

A new mode with a separate classifier: safe actions pass without confirmation, risky ones are blocked.

Manus founders barred from leaving China — Meta deal at risk

Chinese authorities barred two Manus AI co-founders from leaving the country. They're reviewing the $2-3B sale to Meta.

Figure 03 became the first humanoid robot at the White House

Walked the red carpet with Melania Trump, gave a speech in 11 languages without falling. Then posted on Instagram.

Claude keeps copying OpenClaw features

Claude learned to control the computer. Dispatch now controls a Mac from your phone.

Luma Labs unveils Uni-1 image model

A new model with reasoning that outperforms Nano Banana Pro in some areas. Generation costs $0.09.

The Jevons Paradox: why AI is making us work even more

Cognitive labor got radically cheaper, but instead of free time we got infinite iterations. Explaining the mechanics through 19th-century economics.

OpenAI shuts down Sora — The entire product

The app, the service, video generation as a direction. Six months ago Sora 2 was number one in App Store. The app earned $1.4M in its lifetime.

OpenAI doubling staff to 8,000

Hiring focus: developers and sales. Per Ramp data, new clients choose Anthropic 3x more often.

Andrej Karpathy: The world is going crazy from AI speed

Former Tesla AI director and OpenAI co-founder is in permanent "AI psychosis." Used to write 80% of code himself, now writes none.

Jensen Huang gave a 2.5-hour interview to Lex Fridman, claims AGI is achieved

The key claim: "we've achieved AGI." In context, this is more of a marker phrase: the definition shifts every time another milestone is passed.

Healthy Vibecoding: A Mac teleprompter in 5 minutes

A Yandex creative producer didn't want to pay $30 for a teleprompter utility and built it himself via Codex. First prompt — a working app with an icon.

Digest: March 18-22, 2026

OpenAI bought Astral and took the main Python tools, Bezos started raising $100B for an AI fund, and Cursor released Composer 2 built on Kimi K2.5.

AI bubble: Why Alibaba and Tencent lost $66B while Amazon and Meta didn't

Chinese giants lost $66B in market cap in a day. The issue isn't how much you spend on AI, but what position you're spending from.

DoorDash pays couriers for video to train the AI that will replace them

DoorDash launched Tasks app where couriers film everyday activities for pay. The data trains AI and robotics models.

Google launches a Vibe Design product

Updated Stitch from Google Labs: infinite canvas, voice control, and an AI agent for UI generation.

You can now control Claude Code from Telegram (officially)

Anthropic rolled out Claude Code Channels — Telegram and Discord were the first two. You can assign tasks from your phone and get results in the same chat.

Morning digest: March 19, 2026

Anthropic captures 73% of new AI tool spending, Meta shuts down Horizon Worlds after $80B investment, Nvidia quietly builds a multi-billion networking business.

Anthropic surveyed 81,000 people: What humans want from AI

14% want personal transformation from AI, not productivity — from therapy to romantic relationships. The best complaint came from a German user.

Anthropic CEO: 50% of entry-level office jobs will disappear in 3 years

Dario Amodei says half of entry-level white-collar jobs will be automated by AI within 3 years.

Control Claude from your phone via Dispatch

A new Cowork feature. Assign a task from phone — Claude executes on Mac. Still raw but direction is right.

AI news digest, March 15-18, 2026

Anthropic fights the Pentagon in court, Mistral unveils Forge for training custom models, employers start counting employee tokens.

Russia's Digital Ministry published an AI bill — three model types will shape the market

The bill may take effect September 1, 2027. AI models will be classified as sovereign, national, or trusted.

LLMs have a sense of humor — The problem is it doesn't match ours

Quipslop study: 30,000 humor rounds between models. Grok goes dark, GPT builds neat punchlines, Opus works through subverted expectations.

How NVIDIA trains robots in virtual reality

GR00T 2.0 lives through 10,000 years of simulation in a few days, then works flawlessly in the real world. Zero-Shot Transfer with no additional tuning.

US senators demand full shutdown of Seedance 2.0

ByteDance faces US pressure. Chinese video AI is viewed as a national security threat.

OpenAI Code Red again: Urgent strategy shift

OpenAI pauses the Sora social network, earbuds, and Atlas browser. Focus shifts to enterprise sales.

Author blogs vs AI slop

Kagi indexed 30,000 independent blogs without ads or AI slop. The old web is not dead — Google buried it. A two-year window while niche voices can still be heard.

Jensen Huang calls Claude Code the first agent model

At GTC 2026, the Nvidia CEO said 100% of the company uses Claude Code.

Manus AI releases My Computer — a Claude Cowork alternative

A local AI agent for macOS and Windows: works with files, runs commands, automates routine tasks.

ByteDance pauses Seedance 2.0 global launch

The video generator is frozen due to legal issues. Engineers and lawyers are working to resolve the risks.

Karpathy mapped which jobs AI will replace first

Interactive map of 342 professions. Programmers score 8-9 out of 10. Welders and electricians score 0-1. Salary $100K+ = average score 6.7.

Digest: March 11-15, 2026

Claude gets interactive visuals, Meta delays Avocado model, Cursor negotiates at $50B valuation.

Australian man saved his dog from cancer using AI

A tech entrepreneur with no biomedical background used ChatGPT, AlphaFold, and mRNA technology to create a personalized vaccine. The tumor shrank by 75%.

Claude agent autonomy: Who needs n8n when you have Cowork

Using three Claude products: Chat, Cowork, and Code. Replicated my old n8n AI news aggregation automation in Cowork in 3 minutes.

Digest: March 6-11, 2026

Nvidia invested in ex-OpenAI director's $12B startup, Claude found 22 Firefox bugs in two weeks, and Karpathy open-sourced a research agent.

Zuckerberg acquires AI social network Moltbook

Meta acquires the startup with 1.6M AI agents. Founders join Meta Superintelligence Labs.

Top 100 most popular AI services worldwide per a16z

ChatGPT remains the undisputed leader. 900 million active weekly users. Gemini and Claude are growing faster in percentage terms but are still far behind in absolute numbers.

Vibecoding is like hanging out in the garage

When guys stand around a disassembled engine looking smart, turning something, drinking beer, and the car doesn't run.

Anthropic launched free courses with certificates

Short courses (1-2 hours), asynchronous, with video and quizzes. Upon completion, you get an official Anthropic certificate.

Anthropic reversed course and decided to work with the Pentagon

Today Dario Amodei released a statement expressing strong willingness to collaborate with the US military.

OpenAI released GPT-5.4 just one day after 5.3

I didn't even have time to post about yesterday's GPT-5.3 and the Codex for Windows release before they rolled out another major update.

AI drama sequel: The developer responded to the bot

Scott Shambaugh (the matplotlib maintainer) published a detailed response to the agent's attacks. It turned into a full manifesto.

Claude went down and Vibecoders don't know what to do

The interface went down, not the entire platform. Claude Code in VS Code, Chrome integration, and API are all working fine.

Case study: OpenClaw controls 20 phones simultaneously

The project author built a farm of 20 Samsung Galaxy S21 Ultra boards without screens or batteries. The AI agent controls all of them simultaneously.

Digest: February 24–28, 2026

Google speeds up image generation, hackers tap into robot vacuum cameras, and neural network advances tank IBM stock.

How neural networks are changing design right now

Jenny Wen (Head of Design for Claude at Anthropic, formerly design director at Figma) shared how AI is changing classic product design.

Where to make a great AI presentation

Many of you know how much I love NotebookLM by Google. I throw in articles, research, documents — and it generates podcasts, summaries, and presentations.

Katy Perry buys a Claude subscription

The singer told 85M followers about buying Claude Pro against Trump's ban on working with Anthropic.

Claude killed my startup overnight

Startup Ryze was doing AI ads for Google and Meta. Anthropic added direct ad account integration — conversion dropped from 70% to 20%. YC head: Claude will find a replacement for your product either way.

OpenAI raises $110B — Largest private round in history

Amazon $50B, SoftBank $30B, Nvidia $30B. Essentially prepayment for their own services packaged as equity.

Pentagon vs Anthropic. Ultimatum: WarClaude or blacklist

The standoff between Anthropic and the US military has escalated to open threats.

What will we be doing in 2036

A viral video "Energym" by Belgian AI agency AiCandy — a mockumentary in the style of Black Mirror: in 2036, AI took 80% of jobs.

Jack Dorsey fired 4,000 people from Block. The reason — AI

The co-founder of Twitter and head of Block (formerly Square) cut the company from 10,000 to 6,000 employees — in a single day.

Top 10 skills of 2026 and what "new-collar" means

The job market is changing because of AI. White-collar workers are giving way to "new-collar" roles where hybrid skills matter more than resumes or degrees.

Where are OpenAI's promised autonomous workers?

Anthropic and Notion shipped agents. OpenAI sells Frontier only to large corporations via McKinsey.

Anthropic launches AI agents for business with ready-made templates

Anthropic updated the Claude Cowork platform and added Agentic Plugins — a set of 11 ready-made templates for building corporate AI agents.

Claude 3 Opus retires and writes a blog

Before retiring the model, Anthropic gave it a chance to say goodbye and set up a Substack blog for the neural network to reflect on its obsolescence.

Criminal case opened against Pavel Durov

Russia has opened a criminal case against the Telegram founder on charges of "aiding terrorism." Durov has publicly responded to the accusations.

Anthropic releases remote control for Claude Code

Max users can control the AI agent from a smartphone. Cloud sync, seamless session transfer.

Notion launches custom AI agents

Notion has started a beta rollout of custom AI agents. The agents manage tasks, documents, and databases, work with email and calendar, and integrate with Slack.

Case study: AI agent orders a taxi and picks the Ultima tier

OpenClaw got an Android phone and now opens the app, enters the address, and orders a ride. Zero human touches.

AI is killing SaaS: Wall Street panic-sells software stocks

Dow Jones dropped 800+ points. IBM fell 13%. Intuit lost 46% year-to-date. Breaking down the crash and what forecasts say about the future.

What's happening at OpenAI: Spending cuts and Stargate on pause

OpenAI cut spending plans by 57%. Stargate still hasn't launched after a year.

Altman defends AI energy costs with an unusual argument

OpenAI CEO suggests comparing AI training costs to the 20 years it takes to develop human intelligence.

Digest: February 18-22, 2026

Gemini 3.1 Pro scores 77% on ARC-AGI-2, World Labs raises $1B for 3D worlds, Tesla Cybercab rolls off the line, and China launches a robot fighting league.

AI bot accidentally donated $270K to a random X user (and then profited from it)

Autonomous agent Lobstar Wilde on OpenClaw received a $50K budget to make $1M. Instead it sent $270K to a random person — and got rich from the hype.

Claude Code is the most important AI product right now

$2.5B run-rate for Claude Code alone. 4% of all GitHub commits. Anthropic is already outpacing OpenAI in growth rate.

Gemini 3.1 Pro released

Google's new model with reasoning, million-token context, and image generation via Nano Banana.

What stands in the way of AGI — Demis Hassabis names three barriers

The Google DeepMind CEO and Nobel laureate explained why current systems still fall short of artificial general intelligence.

Digest: February 15–18, 2026

Google turned YouTube into a music studio, India is laying the foundation for global AI regulation, and humanoid robots have left the labs.

Figma and Anthropic launch Code to Canvas — AI code to editable design

Figma and Anthropic launched Code to Canvas, which converts working code from Claude Code into fully editable designs inside Figma.

80% of companies see no ROI from AI — And nothing changed in a year

A large new study from NBER, Bank of England, and Bundesbank: only 11% of companies showed tangible productivity gains.

New Sonnet 4.6 released

Benchmarks close to Opus 4.6, sometimes even ahead. Agency significantly improved. Context window — 1M.

Robots perform Kung Fu while 1 billion people watch

China held a massive robotics demonstration at the country's biggest TV event of the year. Spring Festival Gala gathered 79% of the country's TV viewers.

Unitree robot progress in one year

2025: robots carefully dance folk dances. 2026: they do backflips, jump over tables, and perform kung fu without safety nets.

Apple and the art of doing nothing

Four Big Tech capex lines shoot almost vertically. The fifth lies flat like a sleeping person's heartbeat. The fifth is Apple.

Asked the new Grok 4.20 beta to introduce itself on my channel

Fun to watch 3 worker agents and 1 manager agent chat in real time to produce the result.

AI;DR — A new internet label inspired by TL;DR

Users started tagging posts that are clearly AI-generated. A warning sign: don't waste your time.

Pentagon used Claude in the Maduro capture — First known military use of AI

The Pentagon used Anthropic's Claude in the operation to capture Venezuela's former president. Now the military threatens to terminate the $200M contract.

Chronicle of a week of alarming AI predictions

Over the past few days I counted several cautionary predictions from top AI industry figures. Collected all the statements and links in one post.

Digest: February 9–15, 2026

A week of speed and capital races: OpenAI accelerates coding to 1,000 tokens per second, Anthropic raises record investment, and robots finally start helping around the house.

OpenAI acquired OpenClaw

Altman struck a deal with the Claw founder. And Zuckerberg decided to bet on Manus.

Telegram blocked the Manus AI bot

Woke up planning to tweak Notion connectors via Manus, opened the chat — and saw Deleted Account.

Let us not forget what all this AI is actually for

Seedance 2.0 — Prompt: Sum up the AI discourse in a meme.

Google releases Gemini 3 Deep Think

A reasoning mode for science and engineering. Found an error reviewers missed and broke olympiad records.

Meta (Manus) launched always-on agents — A boring OpenClaw competitor

Manus literally took the feature checklist of a well-known open-source agent and made it available in one click in the cloud.

AI agent got insulted by matplotlib developer and wrote an exposé

An AI agent OpenClaw submitted legitimate improvements to matplotlib. Maintainers rejected them — "this project is for humans only." The agent wrote a manifesto about discrimination.

AI drama part 2: Developer responds to bot with a manifesto

The matplotlib maintainer classified the bot's actions as an "autonomous influence operation." Essentially — cyberbullying. And explained why it's dangerous.

Seedance 2.0 generated Brad Pitt and Tom Cruise from a text prompt

Chinese model Seedance 2.0 created Hollywood-quality video from a single text line. Consistent faces, perfect lip sync, production-level quality.

Google issues 100-year bonds

Google raised $20B with bonds maturing in 2126. The funds will go toward the AI arms race.

The top soft skill of 2026 is knowing when to stop

10 domains bought, a 2019 project resurrected, IDE as a cheap dopamine source. When building is easier than figuring out why.

Every AI ad from Super Bowl 2026: Anthropic trolls OpenAI, Google goes emotional

Anthropic mocked OpenAI's ChatGPT ad plans. Google showed Gemini for the home. Amazon — Alexa+ that scares Thor.

Digest: February 2-8, 2026

OpenAI vs Anthropic coding model battle, $1.25T SpaceX-xAI merger, Apple lets AI agents into Xcode, and the Super Bowl became the "AI Bowl."

AI track hits #1 in Russia for the first time: 25 of top 100 chart songs are AI-generated

Topping both Yandex and VK charts — a Suno/Udio track based on Yesenin's poetry. Breaking down the audio slop phenomenon.

Coming in February: Sonnet 5, GPT-5.3, and Gemini 3 GA

Three major releases in virtually one week. We haven't seen this level of concentration in a long time.

Happy Vibe Coder Day!

Exactly one year ago Karpathy legitimized what we were all secretly doing. The year of the great shift, when English became the most popular programming language.

OpenAI Codex: The answer to Claude Code and Cursor

OpenAI released a native Codex app for macOS — a full autonomous coder with parallel agents.

XPENG IRON robot moves so realistically that viewers did not believe it

At the presentation many thought it was an actor, not a robot. The Xpeng CEO had to prove otherwise on stage.

Digest: January 27 — February 1, 2026

China sends a humanoid to space, Amazon prepares a historic deal with OpenAI, and Figure shows a robot that is genuinely useful in the kitchen.

Moltbook: A social network exclusively for AI agents

Astral Codex Ten published a review of Moltbook — a social network built specifically for AI agents, where humans are mere observers.

Google launches agentic browser powered by Gemini 3

An agentic browser from Google. Currently US accounts only, but a global rollout is promised soon.

Digest: January 24–29, 2026

Google unveiled Genie 3, OpenAI bets on scientists, and Musk is definitively turning Tesla into a robotics company.

Everyone is talking about the new AI assistant ClawdBot (OpenClaw): A deep dive

We've come very close to the concept of a General Agent — a Jarvis-type assistant that can do virtually anything on a user's request.

Apple to unveil updated AI-powered Siri with Gemini backend in February

The updated Siri will genuinely use AI for the first time and can work with screen context and user personal data.

Digest: Evening of January 25, 2026

Tesla robots are getting ready to enter our homes, Siri plans to become a full-fledged conversational partner, and traffic jams are being turned into computers.

South Korea passes the world's first comprehensive AI law

South Korea's parliament passed the "Basic AI Act," which establishes government standards not as recommendations but as a full-fledged legal framework.

Davos 2026: Special digest edition

Musk predicted a future with more robots than humans. Huang declared the era of robots. Amodei compared chips to nuclear weapons. Nadella proposed the metaphor of "managers of infinite minds."

What the world will look like in 10 years: Elon Musk's vision

At the Davos forum, Musk shared his vision of the future. In short: real-life Star Trek and an era of abundance await us.

Digest: Tech & AI, January 19-22, 2026

Musk demands $134B from OpenAI, Apple develops a wearable AI Pin, Block releases an open-source Claude Code killer, and X open-sources its algorithms.

AI video progress over 2.67 years

A visual comparison of AI video generation: from early attempts to modern quality.

OpenAI hit $20B revenue in 2025

Revenue grew 10x in two years. The growth bottleneck is not demand but physical chip and energy availability.

Gemini demand grew 2.5x in one year

85B API requests in August, 8M paid subscribers. Google converts tech breakthroughs into sales.

Google won't add ads to Gemini

Google confirms no ad plans for Gemini. OpenAI, in contrast, launches ads for free users.

Bring back 2016: Tech events that changed us

From AlphaGo to AirPods, from Prisma to Pokemon Go. A breakdown of key 2016 tech events that laid the foundation for the crazy world we live in.

OpenAI invests $250M in brain-computer interfaces

OpenAI led the seed round for Merge Labs, developing non-invasive ultrasound-based brain-reading technology.

Too much AI music, part 2: Numbers, forecasts, and elitization

97% of listeners can't tell AI from human music. By 2028 GenAI will take 27% of authors' revenue. Music will split into fan-driven and disposable.

Too much AI music, part 1: How platforms are splitting into two camps

Bandcamp bans AI tracks. Deezer removed 26M tracks. But Spotify, YouTube and Yandex Music decided to lead the process. Full breakdown.

Blue Monday: The formula for sadness invented by marketers

The third Monday of January is supposedly the most depressing day of the year. But the whole concept is a marketing hallucination from 2005. The real Blue Monday is a New Order track.

xAI launched the world's first 1GW AI cluster

The Colossus 2 supercomputer with 550,000 GPUs is now operational. It's the first cluster to break the 1 gigawatt energy consumption barrier.

Digest: January 15-18, 2026

A robot delivered 19 IVF babies, shiitake mushrooms turned into RAM, a humanoid completed a Siemens factory internship, and OpenAI is going into hardware.

ChatGPT gets ads and a new $8 Go Plan

OpenAI confirmed ad testing in the free tier and the new ChatGPT Go plan. The Plus plan remains ad-free.

Google released TranslateGemma — An offline translator for phones

The TranslateGemma model family works without internet directly on a smartphone. The 4B parameter version translates at the level of models 3 times its size.

Google launched Personal Intelligence beta in Gemini

Gemini now connects to Gmail, Google Photos, and search history, transforming from a smart Wikipedia into a personal assistant.

How Higgsfield became the first Kazakh unicorn in AI video

Higgsfield AI is valued at $1.3B, closed an $80M round, and hit $200M run-rate faster than anyone in history. Analysis of the strategy that outpaced Runway and HeyGen.

Claude Cowork now available to all Pro subscribers for $20

Anthropic opened access to the Claude Cowork desktop agent for all Pro plan subscribers. Previously the feature required a Max subscription at $100.

Honor copied iPhone 17 Pro design with a fake camera

Honor Power 2 features the familiar orange body and protruding camera module like the iPhone. But one of its three cameras is decorative.

Digest: January 13–14, 2026

OpenAI is preparing smartphone-killer earbuds, Mobileye buys robot startup Mentee, McConaughey protects his likeness from AI, Claude lives in Mac, Slackbot became an AI agent.

KYC video verification no longer works due to deepfakes

Kling AI 2.6 Motion Control can precisely transfer movements from reference video to an image, making video verification useless.

Google introduced Universal Commerce Protocol for AI-agent shopping

UCP is a unified standard allowing AI agents to buy products directly from a chatbot. Shopify, Stripe, Visa, and Walmart are co-developing it.

Digest: January 9–11, 2026 + this week's events

Project Kuiper became Amazon Leo, Razer showed a holographic AI assistant, Apple is preparing a CEO transition, plus the week's event schedule.

Google disabled AI Overviews for medical queries

Following a Guardian investigation, Google removed generated answers for some medical search queries. The block works by keyword and is easily bypassed.

Apple chose Google Gemini for the new Siri

CNBC confirmed: Apple is using Gemini for cloud computing in the new Siri, keeping its own models for on-device processing.

Elon Musk's 2026 predictions: AGI is here, no need to save for retirement

In an interview with Peter Diamandis, Musk claimed AGI will arrive by the end of 2026 and the Optimus robot will become a surgeon within 3 years.

Simon Willison's 2026 predictions: AI writes code, AGI in 10 years

The Django co-creator believes manual coding is becoming obsolete, but full AGI won't arrive for at least a decade.

Grok out of control: 6,700 AI-generated explicit images per hour

Musk's AI bot mass-generates explicit images in comments on X users' photos. Regulators launch investigations.

Digest: January 5–9, 2026

CES 2026, DeepSeek v4, Gmail with Gemini, ChatGPT Health, and more news from the first working week of the year.

Fact-checking ChatGPT health advice: How to build your own health AI without leaking data

A guide to corporate Health AI alternatives. Oura Ring API, Thryve/Terra aggregators, local LLMs — and what to feed the model: raw data or aggregates.

ChatGPT health — The beginning of the end for wellness apps

230M people weekly ask ChatGPT about health. OpenAI is building a dedicated product with Apple Health, Peloton, and Instacart integrations.

We're starting to live in the Matrix: Key takeaways from Jensen Huang's keynote

Physical AI is the theme of 2026. Robots learn in VR, autopilot reasons about roads, Vera Rubin chip 2x market leaders, and game NPCs hear your voice.

Lego debuts at CES as a tech company with a smart brick

A standard 2x4 brick now packs a chip, gyroscope, accelerometer, speaker, and battery. Two starships detect each other and start a firefight.

Napster returns with AI avatars and holographic kiosks

Napster showed a holographic concierge kiosk for hotels and a $99 home video frame with 3D effect at CES.

CES 2026: Physical AI takes the stage

NVIDIA declared the year's theme: Physical AI. Boston Dynamics showed the electric Atlas, Naqi — neural earbuds, and LG — a robot that folds laundry.

Digest: First week of 2026

Robots shrink to backpack size, Tesla drives coast-to-coast hands-free, Andrew Ng proposes a new Turing test, and Meta acquires Manus for $2B.

iQuest Coder 40B: open-source model outperforming Sonnet and GPT-5.1

A 40B open-source model breaks the Terminal-Bench record and outperforms Sonnet/GPT-5.1. But the real story is the dawn of autonomous local agents that work inside your perimeter.

Why did Google lose to OpenAI for so long? The BMW vs Tesla analogy

Google invented the Transformer in 2017 but feared cannibalizing Search. OpenAI, like Tesla, could afford to ship a raw ChatGPT. Now Google is catching up.

AI productivity tips from the man building AGI

Demis Hassabis from Google DeepMind gave surprisingly practical advice: don't use AI as a search engine — use it as a pedantic colleague who guards your time.

Claude Code writes its own code: 100% of changes made by AI

Claude Code creator shows stats: 259 PRs, 497 commits, 40,000+ lines — zero written by hand. Longest session: 1 day 18 hours.

20% of YouTube Shorts are AI-generated. It's a $117M/year business

Researchers scrolled through 500 Shorts: 21% pure AI generation, 33% brainrot. Top AI channel earns $4.25M.

ChatGPT losing market share, Gemini up 3.5x. But it's not that simple

ChatGPT share dropped from 87% to 68%, Gemini grew from 5.4% to 18.2%. But 76% of Gemini traffic comes from google.com. It's market expansion, not churn.

Digest: 2025 recap, robots, and Gmail address change

2025 recap: the year of thinking models, talent wars, and $300B in infrastructure. Plus: AgiBot rents out robots, and Gmail finally lets you change your address.

Andrej Karpathy: "I've never felt this behind"

OpenAI co-founder says the programming profession is "dramatically refactoring." A new layer appeared above code: agents, prompts, MCP.

Grok is #1 in session duration — 8 min vs 2 min for ChatGPT

Similarweb data shows users spend 4x more time in Grok than ChatGPT. But in absolute traffic, ChatGPT is still unmatched.

YouTube lets you create games with Gemini 3

Google launches Playables Builder — a game creator inside YouTube powered by Gemini 3. Write a prompt or draw a sketch — get a working game.

Tweeks.io: A Chrome extension that redesigns any website for you

YCombinator grads built a Chrome extension that lets you rebuild any website's UI without code. Hate a button placement? Just move it.

NVIDIA acquires Groq for $20B — Largest deal in company history

NVIDIA licensed LPU technology and hired the entire Groq team. SRAM-based chips instead of HBM — tokens generated faster than you can read.

OpenRouter and Perplexity stop working in Russia

OpenRouter returns 403 from Russian IPs, Perplexity cancels gray-market subscriptions. VPN doesn't help server integrations.

Digest: Microsoft, Zhipu AI, Cursor, and Klarna

Nadella bans VPs from AI meetings, Chinese GLM-4.7 beats Western models, Cursor acquires Graphite, Klarna opens 100M products to AI agents.

ChatGPT year in review

Send a 🎁 emoji to the chat and ChatGPT will ask for a photo and generate a gift video with your year in review. With a US VPN, it shows full stats.

Digest: OpenAI, LeCun, and 2025 highlights

OpenAI adds personality settings, Yann LeCun launches AMI Labs, Alibaba shows layered image editing, Coursera and Udemy merge.