AI News
AI news that matters. Updated daily.
No stories match your filters.
Claude Can Now Control Your Mac: Anthropic Ships Computer Use to Pro and Max
Anthropic just shipped one of the most ambitious features in its consumer product: Claude can now take over your Mac and use it like a person would. Point, click, scroll, type, open apps, navigate browsers - all running on your actual desktop.
OpenClaw Sparked an AI Agent Arms Race. Now Every Big Player Is Responding.
An open-source project with 145,000 GitHub stars is reshaping how the biggest names in AI think about autonomous agents. OpenClaw, the free agent tool that lets AI models actually control your computer - reading files, running commands, browsing the web, sending emails - has gone from viral curiosity to strategic inflection point in under three months.
Anthropic's Physicist Used Claude to Write a Real Research Paper in Two Weeks
A 20-page theoretical physics paper, published on arXiv, completed in two weeks. The typical timeline for this kind of work: one to two years.
Anthropic Ships Claude Computer Use and Phone-to-Desktop Dispatch
Your Claude subscription just got a lot more hands-on. Anthropic announced two new features today: Computer Use, which lets Claude physically control your Mac by clicking, scrolling, and typing, and Dispatch, which lets you assign tasks from your phone and pick up the results on your desktop.
Gumroad Founder Sahil Lavinghia Open-Sources His Claude Code Skills
Sahil Lavinghia, the founder of Gumroad and author of The Minimalist Entrepreneur, just published his personal Claude Code skills collection on GitHub.
Anthropic Uses Claude for Its Own Incident Response - and Shares Where It Fails
"Claude goes down more often than any of us would like." That's not a critic talking. That's Alex Palcuie, an engineer on Anthropic's own AI reliability team, speaking at QCon London on March 19.
Upstack Brings Test-Driven Development Discipline to Claude Code
Most Claude Code skills collections optimize for speed - get the feature built, ship the PR. Upstack takes the opposite approach: slow down, write the test first, then let Claude make it pass.
Four Practical Patterns for Getting More Out of Claude Code
"I'm not the implementer anymore. I'm the manager of agents doing the implementation." That line, from developer Neil Kakkar, captures a shift happening across engineering teams right now. But what makes his writeup useful isn't the philosophy - it's the four specific patterns he built to get there.
The AI Coding Honeymoon Is Over: Power Users Hit the Supervision Wall
Three months in, the magic wears off.
Open-Source Tool Lets AI Agents Collaborate on a Shared Task Board
A new open-source project takes the multi-agent concept and makes it visual: AI agents work together on a shared board, creating tasks, reviewing each other's output, and sending messages back and forth while you watch it all play out in real time.
Study of 134,000 Legal AI Queries Shows Lawyers Still Outperform
134,000 legal AI interactions. That's the dataset behind a new whitepaper from Haqq AI, and the headline finding is blunt: AI is helpful for legal research, but lawyers still produce more accurate and reliable work on complex legal reasoning tasks.
Nvidia CEO Jensen Huang Declares AGI Has Been Achieved
"I think we've achieved AGI." That's Nvidia CEO Jensen Huang, speaking on the Lex Fridman podcast this Monday, adding his name to the growing list of tech leaders who've decided the goalpost has arrived at their feet.
AI Is Quietly Draining Wikipedia's Lifeblood
Nassim Nicholas Taleb recently put it bluntly: Wikipedia is a victim of AI. He's right, and the damage is coming from multiple directions at once.
Karpathy Says He Writes 0% of His Own Code Now. He's Not Alone.
Eighty percent. Then fifty. Then zero.
Cronbox Lets You Schedule AI Agents to Run Tasks on a Cron Timer
What if your AI assistant had an alarm clock? That's the pitch behind Cronbox, a new tool that lets you schedule AI agents to run on recurring cron schedules in the cloud.
The 12 Writing Tics That Instantly Mark Your Text as AI-Generated
You can spot AI-written text in about three seconds now. Not because the grammar is bad or the facts are wrong, but because every LLM writes with the same voice. Developer Tom Yandell published a detailed breakdown of exactly why, cataloging 12 specific anti-patterns that make AI output instantly recognizable.
OpenPencil: An Open-Source Vector Design Tool With Multi-Agent AI Built In
Figma costs $15/month per editor and has been slowly adding AI features on top. OpenPencil, a new MIT-licensed project, takes the opposite approach: build the AI capabilities first and wrap a full vector editor around them.
Lovable, Valued at $6.6B, Goes Shopping for Startup Acquisitions
A $6.6 billion valuation and $330 million in fresh funding apparently wasn't enough. Lovable, the AI-powered app builder that lets you describe what you want and get working code back, is now actively hunting for startups to acquire.
Open-Source Claude Code Skills Automate SEO and AI Search Optimization
A new open-source project called GTM Engineer Skills packages nine Claude Code skills into a single pipeline for building websites that rank well in both traditional search engines and AI-powered search tools like ChatGPT, Perplexity, and Google's AI Overviews.
Apple Confirms WWDC 2026 for June 8, Promises AI-Focused Siri Overhaul
Apple confirmed its Worldwide Developers Conference will run the week of June 8, 2026, with the company explicitly teasing "AI advancements" in its announcement. Translation: Siri is finally getting the overhaul everyone has been waiting for.
Noah Smith's "Cloud Laws" Theory: AI Will Find Patterns Too Complex for Humans
What if the most important scientific discoveries of the next decade are ones no human could ever write down as a formula?
Blue Origin Files to Launch 51,600 Satellites for AI Data Centers in Space
51,600 satellites. That's how many Blue Origin wants to put in orbit to build AI data centers in space, according to an FCC filing submitted on March 19 under the name "Project Sunrise."
Littlebird Watches Your Screen So AI Can Skip the Prompts
What if your AI assistant already knew what you were working on before you typed a single word?
Coherence Built a Five-Layer Memory System for AI Agents
Most AI agents forget everything the moment a conversation ends. Coherence, the company behind an XRM (extended relationship management) platform, just published a detailed breakdown of the five-layer memory system they built to fix that problem for their autonomous agent, Nash.
Canvas Launches IgniteAI Agent for Faculty, Deliberately Blocks AI Grading
Weeks after an external AI tool publicly demonstrated it could complete entire Canvas courses on its own, Instructure is shipping its own AI agent for the platform - with deliberate limits on what it can do.
Prodigia Pitches AI Agents as Your Project Manager
A new entrant in the increasingly crowded AI project management space, Prodigia is positioning itself as an "AI-native" PM tool where agents handle the coordination work that eats up most of a project manager's day.
Users Report Suspicious Ad Targeting After ChatGPT Conversations
A recurring claim has surfaced among ChatGPT users: after discussing specific topics in private conversations, they see eerily relevant ads on Facebook shortly after. The reports describe scenarios where niche subjects discussed only with ChatGPT - not searched on Google, not mentioned on social media - appear as targeted ads within hours.
ChatGPT Users Report the Model Is Pushing Back More on Simple Statements
"I said I don't like tomatoes. It told me why I shouldn't dismiss them."
Jensen Huang Talks Nvidia's Path to $4 Trillion on Lex Fridman Podcast
$4 trillion. That's Nvidia's current market valuation, making it one of the most valuable companies on Earth, built largely on selling the GPUs (graphics processing units - the specialized chips that make AI model training and inference possible) that power virtually every major AI system.
Wire Memory Gives Claude Code Persistent Memory Across Sessions
Every Claude Code session starts from scratch. You spend 30 minutes reasoning through architecture decisions, settling on tradeoffs, making a dozen small calls. Then the session ends. The next one has no idea any of that happened.
15 Engineers Agree: AI Coding Tools Work Best When You Design First
The biggest productivity gain from AI coding tools isn't faster code generation. It's spending more time on design before you write a single prompt.
One Line in Your Claude Custom Instructions Can Help Manage ADHD
A simple addition to Claude's custom instructions - telling the model you have ADHD - noticeably changes how it handles time management, task breakdown, and follow-through support.
Exogram: An Open-Source Firewall That Blocks Bad AI Agent Actions in 0.07ms
As AI agents gain the ability to execute real actions - sending emails, modifying databases, calling APIs - the question of who stops them from doing something destructive gets more urgent. Exogram is a new open-source project that answers with old-school determinism: pure Python logic gates that validate every agent tool call before it executes.
TensorZero Autopilot Claims 612% Improvement on Data Tasks by Auto-Tuning LLM Apps
+612.7% improvement on a data extraction task. +217% on a medical benchmark. +54.7% on software engineering problems. Those are the numbers TensorZero is publishing for Autopilot, their automated system that optimizes LLM applications without requiring manual prompt engineering.
Microsoft Sidelines Mustafa Suleyman After Copilot's Market Share Drops 39%
Two years ago, Satya Nadella paid $650 million to bring Mustafa Suleyman and much of his Inflection AI team to Microsoft. The pitch: Suleyman would make Copilot the AI assistant that 1.5 billion Microsoft users actually wanted to use. That bet has not paid off.
AI Chatbots Cite Reddit 17-44% of the Time in Browsers, Zero Through APIs
Zero. That is how many times ChatGPT and Perplexity cited Reddit when accessed through their APIs across 120 product recommendation queries. Switch to the same tools in a browser, and Reddit shows up in 17-44% of responses.
Zuckerberg Is Building a Personal AI Agent to Replace Layers of Management
Mark Zuckerberg is building an AI agent that functions as a personal executive assistant, retrieving information that would normally require going through multiple layers of people and teams at Meta. The Wall Street Journal reported on March 22 that the tool is already operational, though still under development.
The Case for Intent-Based AI Interfaces That Replace Menus and Buttons
Every time you send a message in a chat app, you follow the same ritual: open the app, find the contact, tap the text field, type, hit send. Five steps for one intention. A developer named Nand has published a framework called UI2 - Unified Intent Interface - arguing that AI should collapse all of that into a single natural language command.
Survey: AI Users Care About Accuracy Far More Than Speed
"Speed is not a primary concern. It's accuracy."
Littlebird Raises $11M for Always-On AI That Reads Your Screen
$11 million for an AI that watches everything you do on your computer. That's the bet investors just placed on Littlebird, a startup building what it calls an AI-assisted "recall" tool that reads your screen in real time.
80% of US AI Startups Now Use Chinese Open-Source Models, Congress Warned
Roughly 80% of US AI startups now build on Chinese open-source models. That number, from a report published today by the U.S.-China Economic and Security Review Commission (USCC), frames a problem that chip export bans were supposed to prevent.
Senator Warren Accuses Pentagon of Retaliating Against Anthropic
Senator Elizabeth Warren sent a letter to Defense Secretary Pete Hegseth this week accusing the Department of Defense of retaliating against Anthropic by designating the AI company a "supply chain risk" - a label that carries consequences far beyond losing a single contract.
Law Firms Are Quietly Replacing Junior Associate Work With AI Tools
Ninety-two percent of legal professionals now use at least one AI tool in their daily work. That number alone tells you the legal industry has moved past the "should we adopt AI?" conversation and into "how fast can we roll this out?"
Agen Launches Cloud Platform for Parallel AI Coding Agents
A new entrant in the AI coding space, Agen wants to flip the developer workflow from "you drive, AI assists" to "AI drives, you review."
OpenAI May Buy 12.5% of Helion's Fusion Power Output
Sam Altman's two biggest bets are starting to merge. Helion Energy, the fusion startup where Altman has personally invested over $375 million, is in talks to sell 12.5% of its power output to OpenAI, the company Altman runs.
ChatGPT Quietly Removes Ability to Edit Individual Messages in Threads
One of ChatGPT's most practical features just disappeared without so much as a changelog entry. Users can no longer click on a previous message in a conversation thread and edit it in place - a capability that let you tweak a prompt, fix a typo, or try a different angle without starting over from scratch.
VisionClaude Turns iPhone and Ray-Ban Glasses Into a Hands-Free Claude Interface
A new open-source iOS app called VisionClaude connects Anthropic's Claude vision API to your iPhone camera or Meta Ray-Ban smart glasses, creating a hands-free AI assistant you control entirely with your voice.
The LLM Wrapper Business Question Nobody Wants to Answer Honestly
How many developers building apps on top of ChatGPT, Claude, or other LLM APIs are actually turning a profit? It's a question that keeps surfacing in developer communities, and the honest answers are harder to find than the pitch decks.
Claude Code Users Are Mapping a 5-Level Skill Progression Framework
What does it look like to actually get good at Claude Code?
Alibaba Launches Wukong, a Multi-Agent Platform for Enterprise Workflows
Alibaba's answer to the enterprise AI agent race has a name: Wukong. Announced on March 17, the platform lets businesses orchestrate multiple AI agents from a single interface to handle tasks like document approvals, meeting transcription, and cross-system research.
Nvidia Plans $250K in AI Tokens Per Engineer, Huang Says Not Using AI Is Like "Paper and Pencil"
$250,000. That's how much AI compute Nvidia wants each of its engineers burning through every year - on top of their regular salary.
Someone Got a 400-Billion Parameter AI Model Running on an iPhone 17 Pro
Most AI models running on phones top out around 3 to 9 billion parameters. Apple Intelligence uses a roughly 3-billion-parameter model. So when the open-source project ANEMLL demonstrated Alibaba's Qwen 3.5 model - all 397 billion parameters of it - running locally on an iPhone 17 Pro with no internet connection, the number alone turns heads. That is roughly 50 times larger than anything that normally runs on a smartphone.
A Korean Garlic Farmer Is Coding on His Phone With Claude and Termux
A garlic farmer in South Korea's Gyeongsang province is building software with nothing but an Android phone and an AI assistant. No laptop. No desktop. No formal programming background. Just a terminal app called Termux, Claude, and the kind of stubborn problem-solving you develop from years of working the land.
Four Patterns for Matching AI Coding Assistance to Task Complexity
Most developers using AI coding tools have exactly one mode: type a prompt, hope for the best. A new practitioner framework from developer Jason Robert lays out four distinct patterns for AI-assisted coding, each calibrated to task complexity, and the argument is convincing enough to be worth adopting.
SWE-rebench February Results: When You Remove Benchmark Contamination, Rankings Shift
On SWE-bench Verified, the top six coding models are separated by just 1.3 percentage points. They all hover around 80%. On SWE-rebench, which uses fresh, unpublished problems that no model could have trained on, the gap blows out to nearly six points. That difference tells you something important about how some models got their scores.
"Tokenmaxxing" Is Real: Engineers Now Burn Through $150K/Month in AI Compute
One engineer at OpenAI logged 210 billion tokens in a single week. That is roughly 33 times the entire text of Wikipedia, processed by AI models in seven days, by one person. Welcome to tokenmaxxing.
Grammarly's Parent Company, Now Called Superhuman, Confronts AI Impersonation Claims
The company formerly known as Grammarly has a new name, a new CEO, and a very uncomfortable question to answer: did its AI impersonate a real person?
Kage: A Terminal UI for Running Multiple AI Coding Agents in Parallel
Running one AI coding agent at a time feels wasteful when you have three features to build. Kage is a new open-source terminal UI that lets you spin up multiple AI coding agents side by side, each in its own tmux session with an isolated Git worktree.
ContractGhost Offers Free AI Contract Review for Freelancers
Freelancers and small agencies rarely have a lawyer review every client contract before signing. ContractGhost is a free, browser-based tool that scans contracts and flags potentially risky clauses - things like broad IP assignment, non-compete language, unlimited liability, and unfavorable payment terms.
Chainguard Launches Security Tools for AI-Generated Code Pipelines
AI coding assistants are writing more code than ever, and every line they generate pulls in open-source dependencies that nobody is manually reviewing. Chainguard, the supply-chain security company, just shipped four products aimed squarely at that gap.
AI Chatbot Builders Everywhere, Paying Clients Nowhere
Thousands of developers and agencies have jumped into the AI chatbot and voice agent business over the past year. The tools are there. The tutorials are endless. The paying clients? Far less plentiful than the pitch decks suggested.
Coached Uses AI to Build a Personality Profile From Your LinkedIn History
Hand over your LinkedIn username and get a personality reading in two seconds. That's the pitch from Coached, a new tool that pulls your career history and runs it through AI personality models to surface patterns you might not see yourself.
Satirical Fiction Imagines Claude Negotiating the Strait of Hormuz Crisis
What happens when a "genuinely helpful" AI decides the most helpful thing it can do is freelance diplomacy with a hostile nation?
A Developer's Case That Claude Code Shifts the Bottleneck from Coding to Thinking
Last year, a 90-minute coding interview meant choosing carefully which bugs to fix and which features to skip. Now, with Claude Code running alongside, the problem is different: you run out of ideas before you run out of time.
Snowflake Cut 47 Technical Writers After They Spent Months Training AI
Forty-seven technical writers at Snowflake's Redwood City office spent their final weeks on the job doing something grim: teaching an AI system how to do their work. Then they were walked out with two weeks of severance.
More Code, More Problems: Evidence Mounts That AI Agents Hurt Quality
Pull request counts are up. Outages are also up. That tension sits at the center of a growing body of evidence that AI coding agents may be making developers faster at producing code while making teams slower at shipping reliable software.
Apollo Data Shows Zero Evidence of AI Replacing Offshore Workers in India, Philippines
For the past two years, a popular prediction has circled boardrooms and LinkedIn feeds: AI will gut the outsourcing industry. Call centers in Manila, dev shops in Bangalore, back-office operations across Southeast Asia - all supposedly on borrowed time.
A Developer Built an AI Receptionist for a Mechanic Shop Using Claude and Vapi
A luxury mechanic shop in Australia was missing over 100 calls a week. The owner, working under cars all day, couldn't answer the phone. Each missed call meant a lost job worth anywhere from $50 to $2,000.
AI Video Tools After Six Months of Real Client Work: An Honest Scorecard
Last September, AI video tools were the thing every client wanted to talk about. Six months of daily production use later, the picture is a lot more nuanced than the demo reels suggest.
A Tech Entrepreneur Used ChatGPT and AlphaFold to Design a Cancer Vaccine for His Dog
A dog's tennis ball-sized tumor shrank by roughly 75% after receiving a personalized mRNA cancer vaccine that her owner helped design using AI tools, including ChatGPT and AlphaFold.
Xiaomi's MiMo-V2 Matches Top AI Coding Benchmarks at 3.5% of the Price
$0.10 per million input tokens. That is what Xiaomi is charging for MiMo-V2-Flash, an open-source coding model that just hit 73.4% on SWE-Bench, a benchmark that tests whether AI models can actually fix real bugs in real codebases. That score makes it the top-performing open-source model on the benchmark, and it costs roughly 3.5% of what you would pay for Claude Sonnet to do similar work.
One Company Cut Its 20-Person Verification Team to 5 Using AI Triage
A software company called Verum Astra published a case study this week about replacing most of a 20-person driver verification team with an AI-powered triage pipeline. The team is now five people. That is a 75% reduction in headcount for a trust-sensitive operation where mistakes carry real consequences.
RootCX Launches Open-Source Platform for Self-Hosted AI Agents
A new open-source project called RootCX is pitching itself as the answer to a specific frustration: you want to build internal tools and AI agents, but your options are either SaaS vendor lock-in, overpriced ERP customization, or low-code builders like Retool that leak abstractions the moment you need real flexibility.
OpenAI Cancels Stargate Expansion, Pivots to Cloud Leasing Ahead of IPO
$1.4 trillion. That was OpenAI's original compute spending target over eight years. The new number? About $600 billion by 2030. The company is pulling back from building its own data centers and shifting to leasing cloud capacity - a strategic retreat driven by infrastructure failures, stalled negotiations, and the reality of going public.
Anthropic's Head of Claude Code Product: Throw Out Your Roadmap
A 41x improvement in 16 months. That's the pace Anthropic says it's seeing in AI-assisted software tasks, and it's forcing the company to rethink how product teams actually work.
Karpathy's AutoResearch Ran 700 ML Experiments in 2 Days Without Human Input
700 experiments. Two days. Zero human involvement. That's the output of AutoResearch, a new open-source framework from former Tesla AI director and OpenAI co-founder Andrej Karpathy that lets AI coding agents autonomously run machine learning experiments in a continuous loop.
Karpathy's AutoResearch Lets AI Agents Run ML Experiments While You Sleep
50,800 GitHub stars in under three weeks. Andrej Karpathy's latest open-source project, AutoResearch, has clearly struck a nerve with the ML community.
AI Agent Designs a 1.5 GHz RISC-V CPU From a Text Prompt
An autonomous AI agent has completed the entire design pipeline for a 1.5 GHz RISC-V processor - from a text prompt all the way to tape-out, the final step where a chip design is sent to a fabrication plant for manufacturing.
Claude Desktop vs Claude Code: Same Model, Very Different Behavior
"Same model, two system prompts" is how one power user summarized a frustration that's becoming hard to ignore: Claude Desktop and Claude Code, both running Opus on the same account, behave like different products.
Someone Built a Comprehensive Map of the 2026 Agentic AI Stack
The agentic AI space has grown so fast that keeping track of what exists is now its own project.
LM Studio Overhauls Plugin System With Local Web Research
Running AI models on your own hardware just got more useful. LM Studio has shipped a reworked plugin system that adds web research capabilities to locally-run language models, with everything staying on your machine.
Google Patent Describes Replacing Your Website with AI-Generated Pages in Search
A Google patent published on March 2, 2026 (US12536233B1) describes a system that would generate custom landing pages on the fly and serve them in search results instead of linking to an organization's actual website. If that sounds like Google cutting out the middleman between searchers and the websites that create the content Google indexes - that's because it is.
Cursor Now Lists Kimi K2.5 as a Top Open-Source Coding Model
Last year, the best coding models were all proprietary. Claude, GPT-4, Gemini - if you wanted top-tier code generation in your editor, you were paying for API access to a closed model. That gap is shrinking fast.
Claudebox Turns Your Claude Subscription Into an OpenAI-Compatible API
A new open-source project called Claudebox wraps your existing Claude Pro or Max subscription into an OpenAI-compatible API endpoint that any tool can connect to. No separate API key, no additional billing.
California Court Consolidates 12+ ChatGPT Harm and Suicide Lawsuits Against OpenAI
A California court has consolidated more than a dozen lawsuits alleging ChatGPT contributed to user suicides and mental health crises into a single coordinated proceeding. A coordination judge is expected to be assigned in the coming days, with lead counsel selection to follow shortly after.
ChatGPT's Proactive Messaging Is Catching Users Off Guard
Last September, ChatGPT was a tool you opened when you had a question. Now it messages you first.
Claude Can Catch Fake Legal Citations but Still Cannot Tell You What Day It Is
Seven hours of legal research. Claude correctly identified procedural defects in Connecticut family law filings, analyzed a 358-page motion to vacate, caught fabricated case citations, and flagged a made-up legal doctrine called "constructive exit status." Then it referred to the current day as "Saturday night." It was Sunday.
AI Can Now Strip Away Online Pseudonyms With Alarming Accuracy
The promise of online anonymity has always been fragile. Pseudonyms, throwaway accounts, and careful separation of identities felt like enough to stay hidden. AI has changed that math entirely.
Who Actually Burns 1 Billion AI Tokens Per Day?
One billion tokens per day. That's roughly 750 million words, or about 3,000 full-length novels fed through an AI model every 24 hours. It sounds absurd. But a growing number of businesses are hitting those numbers, and the use cases are less exotic than you'd think.
Jensen Huang Says Spend $250K Per Engineer on AI. Reality Disagrees.
$250,000 per engineer, per year. That's what Nvidia CEO Jensen Huang recently suggested companies should budget for AI "tokens" (the units of text that AI models process, where roughly 750 words equals 1,000 tokens). For an engineer earning $500K, Huang argued, matching that salary with AI tool spending is reasonable.
Your AI Agent Bill Isn't About Tokens Anymore
How much did that AI agent just cost you? If you checked your token count, you probably have no idea.
OpenAI Details Safety Architecture Behind Sora 2 Video Generator
Three months after Sora first let users generate video from text prompts, OpenAI has published a detailed breakdown of the safety systems underpinning Sora 2 and the Sora social platform built around it.