OpenAI is working on a movable home AI speaker with a camera, priced $200–300, from Jony Ive's io Products studio — debut presentation in 2026, sales no earlier than 2027. Apple has filed a lawsuit that could delay the launch.
Apple sued OpenAI for trade secret theft, China caught an orbital rocket stage at sea for the first time, and 200+ economists — including 16 Nobel laureates — warned of sweeping AI-driven job losses.
Siri now holds context, reads the screen, and accesses Mail and Calendar — built jointly with Google on Gemini. Wedbush and JPMorgan predict an upgrade supercycle: 66% of active iPhones are still on iOS 26.
Claude's desktop app now has a built-in browser for autonomous web navigation, and Anthropic extended free Fable 5 access again — now through July 19 in paid plans.
A week of AI price wars: SpaceXAI launched Grok 4.5 at $2/M tokens, Meta opened the Muse Spark API at 4× below competitor rates, and Unitree robots performed the world's first humanoid live surgery.
GPT-5.6 Work brings the agent approach from code and repos to everyday tasks — cloud documents, presentations, and plugins — with no terminal required.
Reflect tracks chat statistics across month, three months, six months, or a year and periodically asks which tasks you want to keep doing yourself even when Claude would be faster.
The AI-2027 team released a sequel — 'Plan A' (AI-2040): open AI research, mutually assured compute destruction between superpowers, and a US-China deal by 2029 to delay superintelligence until 2040.
Nvidia delayed its Kyber rack to 2028, SpaceX launched the first commercial nuclear-powered satellite, and Meta released free AI image generation via Muse Image.
Reuters reports Beijing met with Alibaba, ByteDance, and Z.ai about restricting overseas access to advanced Chinese AI models. The author argues the future belongs to two proprietary stacks — US and China — with no open-source alternative.
Jersey Mike's cited AI in its IPO prospectus, Anthropic cut off the workarounds Chinese firms used to reach its models, and Virginia residents killed a 2,100-acre data center campus.
After three weeks offline, Anthropic restored Fable 5 and Mythos 5 for everyone: the US Commerce Department lifted the restriction, finding no unique risk — the same vulnerabilities could be found by Opus 4.8 and even Haiku 4.5.
The BioShocking attack by LayerX bypasses AI browser guardrails using a logical trap, successfully tested against 6 tools including ChatGPT Atlas and the Claude Chrome extension.
OpenClaw released iOS and Android apps: your phone connects to your server and hands the agent your camera and mic. Cursor also shipped an iOS app, and open-source Orca runs several agents in parallel.
Anthropic's Boris Cherny argues that rigid job titles (frontend, backend, design) are giving way to five archetypes — from "prototyper" to "maintainer" — and strong specialists already cover 2–3 roles at once.
Amid a fuel crisis, marketer Evgeny Chudov built "GdeBENZ" — a crowdsourced gas-station map — with Claude in a few days: 1.8M unique visitors in three days and over 100,000 driver reports.
OpenAI unveiled its first Jalapeño inference chip, the White House slowed GPT-5.6's rollout for a safety review, and Google lost two more Gemini researchers in six days.
Flagship Sol is trained for agentic coding and cybersecurity and supports ultra mode with subagents; priced at $5/$30 per 1M tokens, currently limited preview for select partners.
Nobel laureate John Jumper left DeepMind for Anthropic, Tesla took aim at Nvidia's business with Megapod, and SpaceX launched its Starfall cargo "saucer."
Ahead of its IPO, OpenAI showed up at the ad industry's biggest festival, pitching marketers on ads inside ChatGPT and a Codex tool for building campaigns. In-chat ad revenue for 2026 is estimated at ~$2.4B.
A breakdown of OpenAI's Cannes Lions session: CRO Denise Dresser talked about a shift from a "media operating model" to an "intelligence operating model" — though an insider in the room put it far more bluntly.
Microsoft pulled thousands of engineers off Claude Code over token costs, China's GLM-5.2 topped open models, and NASA handed a Mars mission to a private company for the first time.
Midjourney released its first hardware product — the Midjourney Scanner, built with Butterfly Network: 60 seconds in water produces a 3D image of muscles, bones, and organs at near-MRI quality.
SpaceX acquires Cursor for $60B in stock — for Grok training data. DeepSeek closed its first outside funding round the same day: $7.4B at a valuation above $50B.
Addy Osmani's free Google whitepaper breaks down what separates vibe-coded prototypes from production software — 85% of developers now use AI regularly, with AI generating 41% of new code.
Xiaomi's MiMo Code beat Claude Code on 200-step tasks, Dario Amodei proposed aviation-style AI regulation, and VC admitted a liquidity crisis: 1,500+ unicorns worth $6T on paper, trading at 40–60% discounts on the secondary market.
The US Commerce Department issued an emergency directive; Anthropic fully shut down Fable 5 and Mythos 5 servers for all clients worldwide, citing the technical impossibility of filtering users by citizenship.
SpaceX priced 555.6 million shares at $135 and raised $75B — more than three times Saudi Aramco's 2019 record of $24.9B. Shares jumped to $170 on the first day of trading under ticker SPCX.
The US is exploring an OpenAI equity stake via a sovereign wealth fund, Oracle dropped 11% on AI capex concerns, and Google launched real-time speech translation across 70+ languages.
Bloomberg examined how Anthropic became tech's most coveted employer: up to five rounds, AI banned during interviews, salaries from $250K, and a low cultural-fit score closes every door.
Russia's State Duma introduced fines of up to 700,000 rubles for using foreign login providers. The author argues the law creates data cascade risks and hurts domestic startups far more than Google or Apple.
Apple's AFM 3 Core Advanced is a 20-billion-parameter multimodal model that runs entirely on-device on iPhone, activating only 1–4 billion neurons per query via sparse architecture and NAND flash storage.
From June 23, Fable 5 on Pro, Max, Team, and Enterprise plans will require token credits on top of the subscription — at double the Opus rate. GitHub Copilot ran the same play on June 1.
Lilian Schmidt from Zürich made a TikTok about using ChatGPT as a parenting co-pilot and gathered 27,000 followers in three weeks. She now sells a "Coparent" GPT for $37 — and she's far from the only one.
Apple rebuilt Siri on a custom 1.2-trillion-parameter Gemini model in a deal reportedly costing $1 billion per year; public beta ships with iOS 27 this September.
A breakdown of 8 AI tools at $200–250/month — and the math on whether Claude Max is worth it when Claude Code clocked 3.3M tokens (~$93 at API prices) in one month.
The CEO of Customers Bank ($26B in assets) ran a quarterly analyst call using his AI clone — no one noticed for 30 minutes. Klarna, Zoom, and Meta are all building executive digital twins of their own.
DeepSeek raises its first outside funding at nearly $7B, Microsoft switches Copilot to token-based billing sending some devs from $39 to $3,000/month, and Uber caps AI coding spend at $1,500/month.
A new "Generative AI features" report in Search Console shows how many times your pages appeared in AI Overviews and AI Mode. Impressions only — no clicks, no CTR.
OpenAI is recruiting hardware, ML, and systems engineers to build robots for skilled trades in the near term and personal robots for everyone in the long run.
Uber burned through its annual AI budget in four months and has now capped per-tool spending at $1,500/month per employee — despite having pushed staff to maximize AI use.
The AI bottleneck moved from software to physics — heat and power. Five stagnating industries got a second life: diamonds (+90%), HDDs (+50%), optics (+1148%), nuclear energy, and Toto toilet ceramics (+18% in a day).
Insurance startup Corgi insured 40,000 clients across 49 states by building its entire stack — insurer, reinsurer, and platform — on LLMs, with policies issued in minutes.
Anthropic raised $65B at a $965B valuation and shipped Claude Opus 4.8, Microsoft is building its own coding model, and ByteDance and Mistral are both moving into custom chip design.
Gemini Spark summarizes messages, schedules meetings, and automates routine tasks — it works, but why Google made it a standalone product remains unclear.
Mythos 1 reaches final stage, DeepMind solves 9 Erdős problems for a few hundred dollars of inference each, and Tesla breaks ground on a dedicated Optimus factory targeting 27,000 robots per day.
A consultant's client burned $500M in one month from unlimited Claude access; Microsoft is force-migrating thousands of engineers off Claude Code by June 30, and Uber's CTO admits the company blew its entire 2026 AI budget in four months.
Barclays forecasts 24 million humanoid robots could offset 60% of China's workforce decline by 2035, a pragmatic answer to a demographic crisis no one has solved by boosting birth rates.
The FBI and DHS have formally defined 'anti-tech extremism' following a wave of physical attacks on AI infrastructure, including arson at Equinix and Tesla sites and a Molotov cocktail thrown at Sam Altman's home.
At commencement ceremonies across the US, graduates are booing executives who make AI the centerpiece of their speeches — apparently to everyone's surprise except the graduates'.
DeepMind's agent solved 9 of Erdős's open math problems, Anthropic's Mythos found 10,000+ real-world vulnerabilities, and Spotify is launching licensed AI covers with royalties for original artists.
At Google I/O 2026, Google announced a full merger of AI answers and classic search into a unified interface powered by Gemini 3.5 Flash — and what that means for websites and search ads.
TechCrunch investigated how AI startups inflate ARR: one claimed $50M against $42M in actual filings, another counted unpaid contracts. VCs saw the real numbers but kept quiet.
SpaceX is hiring engineers and physicists for SpaceXAI with no AI background required — just three bullet points proving exceptional ability. The company filed an S-1 last week, with IPO expected in June at a $1.5–2T valuation.
Annual recurring revenue has soared to $45B+ from $9B in a few months; funds like Lightspeed and Dragoneer are ready to pour in up to $50B at a valuation near $1T.
Apple is close to shipping camera-equipped AirPods Pro with Visual Intelligence at $299, but the launch is stalled by an unfinished Siri — a potential slip to 2027.
Salvatore Sanfilippo (antirez), creator of Redis, built ds4 — a program that runs DeepSeek V4 locally on Apple Silicon, with 16k lines written in two weeks alongside GPT-5.5 and 26 tokens/sec on an M3 Max.
Anthropic added /goal to Claude Code — the official take on the ralph loop, a one-liner bash script for running an agent indefinitely with persistent notes on disk.
Starting June 15, Claude Pro, Max 5x, and Max 20x subscribers get separate agent credits of $20, $100, and $200 for use in OpenClaw, GitHub Actions, and Agent SDK; weekly limits also rose 50% through July 13.
A breakdown of Notion's three new developer tools — webhooks, Workers, and CLI — with real-world examples including how AI agents like Claude Code can drive them directly.
A viral OpenClaw demo: an AI agent finds homes without a pool via PropStream, renders one onto a real satellite image, and mails a physical postcard with a QR code — at $1–2 per contact with a potential $50k deal value.
Meta burned 60 trillion tokens in a month over an internal leaderboard called Claudeonomics — worth $900M at Anthropic API rates, over $100M even with discounts. The tokenmaxxing trend has spread to Microsoft and Salesforce, where engineers burn tokens to avoid looking insufficiently AI-native.
OpenAI staffer Roon went viral arguing Anthropic is effectively run by Claude itself — and dissecting why Claude feels like an 'Other' while GPT feels like an extension of yourself.
Anthropic nears a $900B valuation with a $50B round in the works, OpenAI opens Codex to office documents, and Japan Airlines deploys humanoid robots at Haneda airport.
A floating avatar sits above all windows and reflects Codex's current status: working, awaiting review, or errored. Eight defaults exist, plus a skill to generate a custom one.
At AI Ascent 2026, Karpathy said agents now write code faster than humans can read it, and called agentic engineering the next discipline for software teams.
Garry Tan (Y Combinator) weighed in on a thread about AI-era jobs: personal agent designer, second brain engineer, and context editor are already real roles in 2026.
Analyst Ming-Chi Kuo says mass production starts in early 2027, a year ahead of earlier estimates. The device is reportedly screenless, resembling a large brooch with a camera.
Workspace Intelligence automates routine in Docs, Sheets, and Meet, while Chrome Skills remembers personal context — down to your clothing measurements for size matching.
Sam Altman confirmed that ChatGPT subscriptions are now officially permitted in Openclaw and Hermes — previously OpenAI had no stated position on this.
With just $14B in AI capex vs. $125–200B at rivals, Apple posted a record $111.2B quarter. Their strategy: own 2.5B devices and take a 30% cut on every AI subscription sold through the App Store.
Chrome extension Cat Gatekeeper monitors your social media time and covers your entire screen with a ginger cat when the limit runs out — with no dismiss button.
In his opening testimony, Musk revealed the backstory of his falling-out with Larry Page and accused OpenAI's leadership of turning his 'lifeline for humanity' into a profit-driven corporation.
China's NDRC blocked the deal and demanded its unwinding — Manus was considered a key player in the AI agents space, a serious blow to Meta's race with OpenAI.
Anthropic became the first AI startup to cross $1T valuation; the week also brought GPT-5.5, DeepSeek V4, Sony's robot beating elite table tennis players, and Microsoft making Copilot agentic by default in Office.
On the Senior Engineer benchmark, GPT-5.5 scores 62.5 vs. 33.5 for Claude Opus 4.7 — the best OpenAI release in over a year for production code and writing.
Buying the Max plan via App Store or Google Play costs more than on the web; the Pro plan stays at $20 everywhere. Reddit flagged this discrepancy 11 months ago.
On April 20, 2026, all major AI platforms went down simultaneously worldwide — thousands of complaints on Downdetector, OpenAI acknowledged "partial outage" with no ETA. By ~23:40 CEST all services reported operational.
Live artifacts in Cowork mode let Claude connect directly to work apps and generate dashboards with charts that refresh automatically in real time, even days after the chat is closed.
The platform will let users build and host custom AI agents running 24/7, built on a feature called Hermes and tightly integrated with OpenAI's existing Workflows builder.
OpenAI Images 2.0 can render text, generate UI mockups, and pull fresh web data; extended thinking mode lets the model invent its own concept from a short prompt.
The video generation model previously only available in its native interface now works free on both platforms — but after copyright drama it responds cautiously to prompts and reference images.
Researchers argue models train up to 20x slower than possible because 95–99% of backprop signal is destroyed when compressed from token space into internal weight format.
Cambridge philosopher Henry Shevlin joins DeepMind as a staff hire, with a mandate covering machine intelligence, human-AI ethics, and society's readiness for AGI.
The team that built Siri on classical NLP is in a multi-week LLM and prompt-engineering bootcamp — exactly two months before WWDC, where Apple is expected to unveil the updated assistant.
Anthropic released Claude Opus 4.7 amid investor offers valuing the company at $800B; OpenAI expanded Codex into a full computer agent and signed a $20B+ chip deal with Cerebras.
Unitree's R1 humanoid robot is now available globally via AliExpress starting at $6,800. While Tesla and Boston Dynamics sell dozens at $50–150k, Unitree has already moved 5,500 units — mostly to universities and research labs.
Anthropic is preparing to launch Claude Opus 4.7 alongside its own vibe-coding app builder, a direct rival to Lovable — the Swedish startup that hit $400M ARR generating apps from text prompts.
Anthropic launched an updated desktop Claude Code with a built-in terminal and multi-repo support, background Claude Code Routines running on Anthropic infrastructure, and a hidden iOS Assist interface for iterative long-horizon task execution.
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's generative interface now renders interactive elements directly in chat. The author used it to visualize a floor lamp's dimensions before buying.
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.
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.
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.
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.
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 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 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.
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'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.
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 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.
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.
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.
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 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.
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.
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.
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.
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.
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.
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 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 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 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.
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.
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.
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.
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.
Anthropic captures 73% of new AI tool spending, Meta shuts down Horizon Worlds after $80B investment, Nvidia quietly builds a multi-billion networking business.
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.
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.
A tech entrepreneur with no biomedical background used ChatGPT, AlphaFold, and mRNA technology to create a personalized vaccine. The tumor shrank by 75%.
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.
Many of you know how much I love NotebookLM by Google. I throw in articles, research, documents — and it generates podcasts, summaries, and presentations.
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.
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.
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.
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.
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.
The Pentagon used Anthropic's Claude in the operation to capture Venezuela's former president. Now the military threatens to terminate the $200M contract.
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.
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.
The matplotlib maintainer classified the bot's actions as an "autonomous influence operation." Essentially — cyberbullying. And explained why it's dangerous.
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.
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's parliament passed the "Basic AI Act," which establishes government standards not as recommendations but as a full-fledged legal framework.
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."
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.
A robot delivered 19 IVF babies, shiitake mushrooms turned into RAM, a humanoid completed a Siemens factory internship, and OpenAI is going into hardware.
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.
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.
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.
Following a Guardian investigation, Google removed generated answers for some medical search queries. The block works by keyword and is easily bypassed.
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.
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.
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.
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.
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.