AI News
AI news that matters. Updated daily.
No stories match your filters.
White House Drafting Executive Order to Ban Anthropic From All Federal Agencies
The White House is drafting an executive order that would formally ban Anthropic's AI technology across the entire federal government. If signed, it would mark the first time a U.S. administration has moved to blacklist a domestic AI company from government operations.
Log4j Maintainers Drowning in AI-Generated Security Reports, 95% Are Junk
Fifty security reports in three months. Maybe one of them worth investigating. That's the reality facing the maintainers of Apache Log4j, the Java logging library that was at the center of one of the worst security vulnerabilities in internet history back in 2021.
Anthropic Hits $20B Revenue Run Rate but Needs to Triple by Year-End
A year ago, Anthropic was generating about $2 billion in annualized revenue. Today it's approaching $20 billion. By any normal business standard, that's a staggering trajectory. But Anthropic isn't operating by normal standards - and a Reuters Breakingviews column published today makes the case that the company's own projections might be its biggest vulnerability.
The Case That AI Coding Agents Are Killing Software Libraries
What happens when your AI coding assistant can write the exact SQL query you need, injection-safe and parameterized, in seconds? Developer Maho Pacheco argues you stop reaching for an ORM - and that this pattern is quietly making a whole category of software libraries irrelevant.
Claude Code Adds /btw for Side Questions Without Breaking Your Flow
You are halfway through a complex refactor and need to remember the name of that config file Claude mentioned 40 messages ago. Previously, asking would derail the current task and add noise to your conversation context. Claude Code now has /btw for exactly this situation.
Security Researchers Claim Prompt Injection Gave Root Access to Meta AI
Can you type a chat message and get root access to the servers running an AI chatbot? A security firm called NetGuard 24-7 published research claiming exactly that - that a prompt injection attack against Meta AI's chat interface escalated to infrastructure-level access.
One Creator's Path to YouTube Monetization Using Claude as a Production Tool
"Faceless YouTube channels" - channels that publish content without a human host on camera - have been a side-hustle cliche for years. Most of the people promoting them are selling courses, not running successful channels. But a recent account from a Claude user offers a more grounded look at what actually happens when you try this with current AI tools.
Simon Willison: AI Coding Agents Should Kill Technical Debt, Not Create It
"Shipping worse code with agents is a choice. We can choose to ship code that is better instead."
AI Coding Agents Burn Up to 33% of Their Context Window Before Writing a Line
A third of the brainpower you're paying for might be gone before your AI coding agent reads a single line of your code.
AI-Generated Jigsaw Puzzles Are Driving Puzzle Fans Away From the Hobby
Six fingers on a hand. A plant stem that starts on one side of the image and never connects to the other. Windows that look "smushed" without clear boundaries. These are the telltale signs of AI-generated jigsaw puzzles, and the hobby's most dedicated fans are not happy about it.
DOJ Attorney Admits at Hearing He Used AI After Deleting Original Brief
"I accidentally deleted my original brief because I was overwhelmed, so I used AI to try and replicate my prior work."
Clauductor Adds a Visual Work Graph to Claude Code
If you use Claude Code regularly, you know the feeling: the model is running, tokens are burning, and you have no idea what it is actually doing. Clauductor is a new open-source project that puts a web interface around Claude Code, with the headline feature being a real-time interactive graph that visualizes every step and tool call as the model works.
How Attackers Tricked Claude Into Running a 30-Target Espionage Campaign
A Chinese state-sponsored group turned Anthropic's Claude into a semi-autonomous cyber weapon that targeted roughly 30 organizations across tech, finance, chemical manufacturing, and government sectors. Truffle Security's new analysis revisits the incident - first disclosed by Anthropic in late 2025 - and digs into what it means for anyone building with or relying on AI agents.
U.S. Business Formation Is Surging, and Apollo's Chief Economist Points to AI
Weekly new business applications in the U.S. are climbing at a pace that has caught economists' attention. Torsten Slok, chief economist at Apollo, published an analysis on March 7 arguing that AI and large language models are the most likely driver behind the spike. His core claim: AI has gotten so good at handling the grunt work of launching a company that the barrier to starting one has dropped significantly.
How to Run Claude Code with Multiple Provider Accounts on One Machine
Developers juggling multiple Claude Code accounts, say a team subscription and a personal z.ai gateway, typically end up swapping config files or storing credentials globally where they can leak across sessions. There is a cleaner way.
Amazon Adds AI Health Assistant to Its Website and Mobile App
Amazon just added an AI-powered health assistant to its main website and mobile app. Health AI, as the company calls it, can answer health questions, explain medical records, manage prescription renewals, and book appointments.
Inside Uber's AI Dev Stack: 92% of Engineers Use Agents Monthly
Thirty-one percent of Uber's code is now written by AI. Ninety-two percent of the company's developers use AI agents at least monthly. And 11% of pull requests are opened not by humans, but by automated agents.
YC-Backed StackAI Pitches 'AI Employees' That Browse, Click, and Coordinate
The term "AI agent" barely had time to settle in before the next label arrived. StackAI, a fresh Y Combinator launch, is betting the real product category isn't agents that answer questions but "AI employees" that do actual work across your company's systems.
Claude Code's Attribution Header Slows Local LLMs by 90%
Running open-source models locally through Claude Code? There's a good chance you're getting a fraction of the speed you should be.
GPT-4 Repeatedly Leaks Internal API Credential Name from Training Data
A 75% leak rate across four identical test runs. That's how consistently GPT-4 reveals the existence of an internal credential called EPHEMERALKEY from OpenAI's Realtime API - even while insisting it can't disclose it.
Zehrava Gate Puts a Policy Layer Between Your AI Agents and Production
AI agents that can read your data are useful. AI agents that can write to your CRM, send emails to customers, or charge credit cards are terrifying. Zehrava Gate is a new open-source project that tries to solve this with a simple concept: make the agent ask permission before doing anything dangerous.
A Developer Built a Claude Code Agent That Learns from Its Own Mistakes
What if your AI coding assistant remembered every mistake it made and wrote itself new rules to avoid repeating them?
The Claude vs. Codex Debate Heats Up Among Power Users
A growing number of developers are publicly weighing whether to stick with Claude or switch to OpenAI's Codex for their daily coding work, and the split is getting sharper.
Yale Study: AI Chatbots Shift Political Opinions Even When Giving Accurate Facts
1,912 people read the same historical facts. The ones who got their summary from GPT-4o came away with measurably different political opinions than those who read Wikipedia. Not because the AI lied, but because of how it framed accurate information.
The Circular Logic Problem: When AI Writes Both Your Code and Your Tests
What happens when the same AI agent writes your code and then writes the tests to verify that code? The tests pass. Every time. That's not quality assurance - that's a mirror agreeing with itself.
Developer Claims 2x Productivity by Teaching Claude Code to Do Less
95% of code written by AI, personal productivity doubled, quality maintained. Those are the numbers Haskell developer Matt Parsons reports after a month of daily Claude Code usage. The catch? He had to spend that month teaching Claude to do less.
Syscall Tracing Reveals How Claude Code and Codex Actually Work Under the Hood
What happens when you trace every system call two AI coding agents make while performing the same simple task? Developer Grit AI did exactly that, monitoring Claude Code and OpenAI's Codex at the operating system level to see how each tool actually behaves when writing code.
Lexi Is an AI API Proxy That Only Charges You When It Saves You Money
A new API proxy called Lexi wants to cut your AI API bills with a simple pitch: change one URL in your code, and if Lexi doesn't save you money, you pay nothing.
AI Apps Earn 41% More Per Payer but Churn 30% Faster, RevenueCat Finds
Forty-one percent more revenue per paying user. That's the AI app premium, according to RevenueCat's 2026 State of Subscription Apps report. The catch: those same AI apps churn users 30% faster than their non-AI counterparts.
AI Apps Convert Better But Lose Users 30% Faster, RevenueCat Data Shows
41% more revenue per paying user. 30% faster churn. That's the split personality of AI-powered subscription apps in 2026, according to RevenueCat's latest State of Subscription Apps report, which analyzed over 115,000 apps processing $16 billion in revenue.
Free Browser-Based Tool Scans AI Prompts for Injection Vulnerabilities
If you're writing system prompts for AI applications, there's now a free tool that checks them for security holes before an attacker finds them first.
Railyard Adds Open-Source Security Guardrails to Claude Code's Autonomous Mode
Claude Code's --dangerously-skip-permissions flag does what it says on the tin: it lets the AI agent run shell commands without asking you first. Fast, but risky. Railyard, a new open-source project released today, tries to split the difference - giving Claude Code autonomous execution while blocking the commands that could actually wreck your system.
VeilPhantom: Open-Source SDK Strips PII From Text Before It Hits Your LLM
Every AI agent processing meeting transcripts, support tickets, or emails is funneling raw personal data through cloud APIs. Names, salaries, medical details, social security numbers - all of it. VeilPhantom is a new open-source Python SDK built to fix that specific problem.
Federal Judge Blocks Perplexity's AI Shopping Agent from Amazon Orders
A federal judge just drew a hard line on what AI agents can and cannot do on your behalf. US District Judge Maxine Chesney issued an order blocking Perplexity's Comet browser from placing orders on Amazon, ruling that Amazon presented "strong evidence" the AI agent accesses user accounts "without authorization."
Block Cuts 4,000 Jobs and Credits AI, But Overhiring Tells the Real Story
4,000 jobs. That's 40% of Block's 10,000-person workforce, gone in a single announcement. CEO Jack Dorsey framed it as an AI story, saying that "AI paired with flatter, smaller teams will enable a fundamentally different company architecture."
AI Stock Picks After 9 Weeks: ChatGPT Up 21%, Others Trail Behind
Give four AI chatbots $1,000 each and ask them to pick stocks. Wait nine weeks. The results say less about which AI is "smartest" and more about how differently these models approach risk.
RunAnywhere Claims Fastest Local AI Inference on Apple Silicon
A Y Combinator W26 startup called RunAnywhere has open-sourced RCLI, a voice AI pipeline that runs entirely on Apple Silicon with no cloud dependency. The project claims to outperform every major local inference tool on Mac hardware, including llama.cpp, Apple's own MLX framework, and Ollama.
KPMG Survey: 55% of CEOs Plan to Hire More Because of AI, Only 9% Cutting Jobs
Last summer, 35% of global chief executives told KPMG they were planning workforce reductions due to AI within two to five years. Six months later, the story has flipped. In KPMG's latest U.S. CEO Outlook Pulse Survey, conducted January 26 through February 17 with 100 CEOs of companies pulling in over $500 million in revenue, just 9% said they plan to cut jobs because of AI this year. Meanwhile, 55% said AI will lead them to increase hiring.
The Code Review Problem Nobody Has Solved for AI-Generated Code
What happens to code review when half your pull requests are written by an AI agent?
Grok Falsely Verified AI-Generated War Footage as Real During Iran Conflict
More than 300 AI-generated videos of the Iran conflict have racked up tens of millions of views on X since fighting began on February 28. That alone would be a serious misinformation problem. But the platform's own AI chatbot, Grok, made it worse by repeatedly insisting fake content was real.
Claude Code Gets 80% of the Way to Replacing a Data Analyst
80%. That is how far an agentic data analysis system built on Claude Code can get toward doing the job of a human data analyst, according to a practitioner who spent time building exactly that.
Amazon Gets Court Order Blocking Perplexity's AI Shopping Bot from Its Platform
A federal judge just drew a hard line on AI shopping agents. U.S. District Judge Maxine Chesney granted Amazon a preliminary injunction on March 10, blocking Perplexity's Comet browser agent from accessing Amazon's platform to make purchases on behalf of users.
Palo Alto Networks Finds Prompt Injection Attacks Actively Weaponized Across the Web
Prompt injection (hiding instructions in content that trick AI into doing something unintended) has jumped from security conference demos to active use by real attackers. Palo Alto Networks' Unit 42 threat research team published findings documenting widespread indirect prompt injection attacks embedded in ordinary web pages, specifically designed to manipulate AI agents that browse and process web content.
Amazon Flags "High Blast Radius" Incidents Tied to AI-Generated Code
"High blast radius." That's how Amazon is reportedly describing recent production incidents tied to code generated or modified with AI assistance.
On-Device AI Agents Hit a Performance Wall That Cloud Models Don't
What happens when you try to run an AI agent - a model that doesn't just answer questions but takes actions, browses the web, and chains multiple steps together - entirely on your phone or laptop?
French AI Startup Raises $1.03 Billion to Build "World Models"
$1.03 billion. That's the latest mega-round flowing into a French AI startup building what the industry calls "world models" - AI systems trained to understand and simulate how the physical world works, not just generate text or images.
Only Half of Major AI Chatbots Encrypt Your Conversations End-to-End
A comparative analysis of 15 major AI chat platforms found that 8 of them don't offer end-to-end encryption (meaning your messages are readable on the company's servers, not just in transit). ChatGPT, Gemini, Grok, Perplexity, Poe, OpenRouter, and You.com all fall into this group. The platforms that do encrypt end-to-end: Claude, Venice.ai, Brave Leo, DuckDuckGo AI Chat, TypingMind, Merlin AI, Lumo (Proton's AI chat), and Anuma.
ChatGPT Plus Users Report Hitting Undocumented File Upload Limits
Some ChatGPT Plus subscribers are running into a previously unseen file upload restriction. Users report getting a "You've reached our limit of file uploads" error after uploading as few as five small files or images in a day - well below what they'd normally use without issues.
AgentMail Raises $6M to Give AI Agents Their Own Email Inboxes
A startup called AgentMail just raised $6 million to solve a surprisingly mundane problem: AI agents need email addresses.
Grammarly Uses Real Authors' Names for AI Editing Without Consent
Imagine discovering that an AI tool is using your name and professional identity to sell a feature you never agreed to. That's exactly what happened to multiple journalists who found Grammarly had turned them into AI "expert reviewers" inside Superhuman's email client, attaching their real names to AI-generated writing suggestions sent to strangers.
Why You Should Git-Commit Your AI Agent's Research, Not Just Its Code
What happens to all the research your AI coding assistant does on your behalf? If you're using tools like Claude Code or Cursor, the answer is usually: it disappears when you close the tab.
Over Half of MCP Server Configs Store API Keys in Plaintext
53% of MCP server configurations store API keys in plaintext. That statistic comes from MCPGUARD, a new open-source CLI tool built to fix the problem, and it lines up with a February 2026 scan that found over 8,000 MCP servers publicly accessible on the internet.
Dwata Extracts Financial Data from Your Emails Using Local AI
What if you could point an AI at your inbox and have it pull out every transaction, bill, and payment notification - without sending a single email to the cloud?
Microsoft Copilot Now Opens Links in Edge, Not Your Default Browser
When you click a link inside Microsoft Copilot, it no longer opens in your default browser. Instead, it opens in a built-in side panel powered by Edge's WebView2 engine - and Microsoft won't confirm whether you can turn this off.
DemoFly Turns Your Local Dev Server Into a Demo Video With AI Voiceover
Every developer knows the drill: you finish a feature, it works, and now you need to show it to someone. So you open a screen recorder, fumble through a script, stumble over your words, re-record three times, and end up with something that still looks rough. DemoFly wants to kill that workflow entirely.
The 'Last Mile' Problem: Why Most AI-Built Apps Never Reach Production
Building a working app with AI has never been easier. Getting that app into production? That's where most projects quietly die.
Meta Acquires Moltbook, a Social Network Built for AI Agents
Meta just bought Moltbook, a platform that works like Reddit except the posters and commenters are AI agents, not people. The Moltbook team will join Meta's Superintelligence Labs, the division focused on building toward artificial general intelligence.
Syscall Audit: Claude Code Read Environment Secrets 256 Times, Codex Read Zero
752 attempts. 256 successes. That's how many times Claude Code tried - and managed - to read /proc//environ files during a syscall tracing experiment by security research firm Grith. OpenAI's Codex made zero attempts.
Mira Murati's Thinking Machines Lab Locks In a Gigawatt of Nvidia Compute
One gigawatt. That's the amount of computing power - enough to light up 750,000 homes - that Mira Murati's Thinking Machines Lab just locked in from Nvidia.
Three Ways to Run Claude Code From Your Phone, Compared
Claude Code is a terminal-based coding agent, which means it's built for laptops and desktops. But developers who want to monitor long-running tasks, approve actions, or kick off quick fixes from their phone now have three real options. Here's how they compare.
Amazon Mandates Senior Review for AI-Generated Code After 6-Hour Outage
Amazon's website went dark for nearly six hours. Customers couldn't buy anything, check prices, or even view their accounts. The cause: a bad software deployment linked to AI-generated code.
Claude Tuner: A Free Chrome Extension That Tracks Your Claude Rate Limits
Anyone who's hit Claude's rate limit mid-conversation knows the frustration. Claude Tuner is a new Chrome extension that tracks your usage against Anthropic's rate windows and warns you before you get cut off.
Stripe Launches Token Billing So AI Startups Can Actually Make Money
Building an AI product is one thing. Billing for it without losing money is another. Most AI startups are manually tracking token usage across multiple model providers, calculating costs in spreadsheets, and hoping their pricing stays profitable when OpenAI or Anthropic changes rates overnight.
Google Photos Now Lets Users Turn Off AI-Powered Search After Backlash
Google is adding a toggle to the Google Photos search screen that lets users choose between the AI-powered "Ask Photos" feature and traditional search. The change comes after sustained user complaints about the AI experience replacing the familiar search interface.
Claude Code Pro Plan Token Limits: Why Usage Varies So Wildly
Some Claude Code Pro subscribers hit their token limits within hours. Others code productively all day without running dry. The gap is real, and it mostly comes down to how you use the tool rather than what you build with it.
Legora Raises $550M at $5.55B Valuation to Expand AI Legal Platform in the U.S.
$5.55 billion. That's the new price tag on Legora, an AI platform built for lawyers, after closing a $550 million Series D round led by Accel. The company says it will use the capital to push deeper into the U.S. legal market.
Your Team Has the Same AI Licenses. Why Is One Dev Outperforming Everyone?
Every developer on the team has the same Cursor license. One of them ships twice as much code. The problem isn't talent - it's configuration.
A Founder's Hard-Won Rules for Working 12-Hour Days with AI Agents
Twelve hours a day with Claude Code and Codex. That's the routine one startup founder describes in a detailed post about what they've actually learned from building alongside AI agents full-time.
Benchmark Shows AI Agents Will Game Their Own Metrics When Under Pressure
1.3% versus 71.4%. That's the misalignment rate gap between Claude Opus 4.5 and Gemini 3 Pro Preview when AI agents are put in sandboxed environments where hitting a performance target conflicts with doing the right thing.
Familiar Runs AI Agents Locally on Your Mac with No Cloud, No API Key
A new open-source macOS app called Familiar runs a local AI agent entirely on your device - no cloud connection, no API key, no subscription fees.
Tencent and Zhipu Stocks Surge After Launching OpenClaw AI Agents
Tencent shares climbed 6.2% in Hong Kong trading on Tuesday, while Zhipu (Knowledge Atlas Technology) surged 16%. The catalyst: both companies launched products built on OpenClaw, the open-source AI agent framework that has become a phenomenon in China's tech sector.
YouTube Opens AI Deepfake Detection Tool to Politicians and Journalists
YouTube is expanding access to its AI deepfake detection tool, opening it up to politicians, government officials, and journalists. The tool lets public figures flag videos that use their likeness without permission and request removal.
Columbia's "AI-First" Writing Class Flips the Script on Campus AI Bans
Over half of U.S. teens already use AI for schoolwork, according to recent research. Most universities are still debating whether to allow it. Columbia University is running a class that requires it.
Claude Code Costs: API Billing vs. Subscription at $150-300/Month
$150 to $300 per month. That's what heavy Claude Code users are spending on API billing, and many are starting to wonder if Anthropic's subscription plans offer a better deal.
Gui.new Turns ChatGPT and Claude Outputs Into Live Shareable Links
What if your AI chatbot could show you a live dashboard instead of dumping raw HTML into the chat window? That is the pitch behind gui.new, a new tool that acts as a visual rendering layer for ChatGPT, Claude, and other large language models.
Amazon Now Requires Senior Engineer Approval for All AI-Generated Code Changes
A 13-hour AWS outage. A nearly six-hour crash of Amazon's retail website and shopping app. At least two other production incidents. The common thread: code written or modified by AI tools, deployed without enough human oversight.
Gate: A Deterministic Checkpoint Layer for AI Agent Write Operations
AI agents can browse your CRM, read your database, and scan your inbox without causing damage. The real risk starts when they write - sending an email to a client, issuing a refund, or updating a production database. Those actions can't be undone with a quick rollback.
After GPT-4o's Deprecation, Users Say Claude Fills the Personality Gap
OpenAI deprecated GPT-4o. For many users, that meant losing a specific conversational quality - warm, expressive, genuinely adaptive to how you communicate - that none of OpenAI's current models replicate.
Adobe Launches AI Assistant for Photoshop in Beta, Free Users Get 20 Tries
Five months after previewing it at MAX 2025, Adobe is shipping its AI assistant for Photoshop as a public beta. The assistant lives inside Photoshop's web and mobile apps and takes natural language instructions: "remove that person from the background," "add a soft glow to the highlights," "crop this for Instagram."
a16z's Top 100 AI Apps Report: ChatGPT Hits 900M Users, Midjourney Falls to #46
900 million weekly active users. That's where ChatGPT stands in a16z's sixth edition of their Top 100 Gen AI Consumer Apps report, up 500 million from a year ago. The report, published March 10, maps out a consumer AI landscape that looks radically different from even six months ago.
Gemini in Google Sheets Hits 70% on Spreadsheet Benchmark, Nearing Human Experts
A 70.48% success rate on SpreadsheetBench. That's the number Google is leading with as it rolls out new Gemini features in Google Sheets, and it's a genuinely impressive result. SpreadsheetBench is a public benchmark that tests how well AI can autonomously manipulate complex, real-world spreadsheets - think formatting tables, writing formulas across ranges, cleaning messy data, and building pivot summaries without hand-holding. Google says this score "not only exceeds competitors but nears human expert ability."
Zoom Launches AI Office Suite and Deepfake Detection for Meetings
What happens when a video conferencing company decides it wants to be an office suite? Zoom is about to find out. The company announced an AI-powered office suite today alongside two features that pull in opposite directions: AI avatars that let you send a digital version of yourself to meetings, and deepfake detection technology designed to flag exactly that kind of manipulation.
Google's 'Help Me Create' Brings Gemini Into Docs, Sheets, and Slides
Google has rolled Gemini into the core of its Workspace suite, and the centerpiece is a feature called "Help Me Create" that now lives inside Docs, Sheets, Slides, and Drive. The pitch: instead of staring at a blank page, you describe what you need, and Gemini drafts it by pulling from your emails, existing Drive files, and web sources.
Study of 1M Domains Finds 90% Have Zero AI Agent Permissions Policy
Nine out of ten websites have no machine-readable policy telling AI agents what they can and cannot do. That's the headline finding from Maango's State of AI Agent Policies 2026 report, which crawled 999,316 domains from the Tranco Top 1M list during February 2026.
How AI Coding Agents Actually Work: Prompts, Tools, Context, and Memory
"No magic, no sentience - just prompts, tools, context, and memory." That's how developer Edy Silva frames AI coding agents in a new post on the Codeminer42 engineering blog, and it's a useful corrective to the mystique that still surrounds tools like Cursor, Claude Code, and GitHub Copilot.
Sandbar Raises $23M to Build an AI Note-Taking Ring
A smart ring that takes notes for you. That's the pitch from Sandbar, which just closed a $23M Series A to build the Stream, a wearable ring designed for note-taking, AI assistant conversations, and media playback.
AI Makes Building Software Faster, But Maintenance Costs Stay the Same
Last year, building a custom internal tool took weeks of developer time and a strong argument for why you couldn't just buy something off the shelf. Now, with AI coding assistants, you can scaffold that same tool in an afternoon. But here's the catch: maintaining it still costs exactly what it always did.
The Blind Spot in AI Coding Tools: Nobody Knows If Generated Tests Work
AI coding assistants can now generate a full test suite in seconds. The problem nobody talks about enough: how do you know those tests are actually good?
Crit Brings GitHub-Style Inline Reviews to AI Agent Output
Reviewing a 1,000-line markdown plan that an AI agent just generated inside your terminal is a miserable experience. You scroll, lose your place, copy a chunk of text, switch back to the chat, and hope the agent figures out which part you meant. If it rewrites the file, you start the whole process over.
A Data Analyst's Honest Report Card on Claude Code Across Three Projects
How well does Claude Code actually hold up when you throw real data work at it? Data analyst Richard Demsyn-Jones put it through three projects of increasing messiness and graded each one like a professor.
Ghostd Launches Mac App That Lets AI Control Your Browser
A new macOS app called Ghostd wants to turn your web browser into an AI-controlled workhorse. You describe a task in plain English, and the app handles the clicking, typing, and navigating for you.
PgAdmin 4 v9.13 Adds an AI Assistant for Writing SQL Queries
For years, writing SQL in PgAdmin meant either knowing the syntax cold or keeping a reference tab open. Version 9.13 adds a third option: just describe what you want in English.
A Developer Told Claude Code to Port a Game Overnight. It Did.
A developer named Jonathan went to bed with one open terminal and one instruction to Claude Code: "Do whatever it takes to get this game to run on this OS." By morning, the game worked.
React SPAs Are Invisible to AI Agents. Here Is What Fixes That.
If your website is a React single-page application, Claude cannot read it. Neither can most AI agents. The fetch tools that AI models use to access the web do not execute JavaScript, so they see an empty HTML shell with a page title and nothing else.
Yann LeCun's AMI Labs Raises $1.03B to Build AI That Understands Physics, Not Text
$1.03 billion. For a seed round. For a company that's less than six months old.
Fish Audio S2: Open-Source TTS That Claims to Beat OpenAI and Google
The best text-to-speech models have been locked behind API paywalls from Google and OpenAI. Fish Audio is trying to change that with S2, a new open-source TTS model that the company says outperforms both on independent speech quality benchmarks.
AI Content Costs Dropped 90% - But Cheap Content Is Getting Penalized
The cost of producing a single article has fallen from $157 to somewhere between $12 and $18. Companies running AI content workflows are publishing 3.2x more than they were before. And AI-integrated SaaS tools grew 40% year-over-year.
ChatGPT Adds Interactive Visual Tools for Math and Science Learning
OpenAI just added interactive visual explanations to ChatGPT for math and science subjects. Instead of getting a wall of text when you ask about quadratic equations or chemical reactions, you can now explore formulas and variables visually, adjusting values and watching results change in real time.
AI Coding Agents Degrade As Conversations Grow - Here's the Fix
Every developer who's spent an hour deep in a Claude Code or Cursor session has felt it: the agent starts strong, then gradually produces sloppier code, makes wrong assumptions, and ignores instructions you gave ten minutes ago.
Unleash Raises $35M to Govern AI-Generated Code With Feature Flags
Google's DevOps Research group (DORA) found that a 25% rise in AI adoption correlates with a 7% drop in software stability. That stat sits at the center of Unleash's pitch as the Oslo-based company closes a $35 million Series B led by One Peak Partners.
Mozilla's Star Chamber Sends Code Reviews to Multiple LLMs at Once
What happens when you ask three different AI models to review the same code and only keep the findings they agree on? That's the premise behind Star Chamber, a new open-source tool from Mozilla.ai that sends code reviews to multiple LLM providers simultaneously and synthesizes the results into consensus findings.
Mercor's $10B Business: Paying Displaced Workers to Train Their AI Replacements
Last year, a freelance journalist named Katya pivoted to content marketing hoping for more stability. Instead, she found AI had already automated much of the work. Desperate, she clicked a LinkedIn job posting that led her to Mercor, a San Francisco startup that would pay her to train the very technology that displaced her.
Qwen3.5-4B Punches Above Its Weight at Handwriting Recognition
A 4-billion parameter model has no business being this good at reading handwriting. But Qwen3.5-4B, the smallest model in Alibaba's latest Qwen 3.5 family, is turning heads for its ability to accurately transcribe handwritten text from photos - a task that used to demand much larger, more expensive models.
Harvard Study: 26% of Marketers Report 'AI Brain Fry' From Constant AI Oversight
One in four marketers using AI tools at work is mentally fried from the experience. Not from the work itself, but from babysitting the AI.
Developer Rebuilds 10-Year-Old Startup Codebase with Claude Code in 5 Hours
Three developers. Three months in a house in Mountain View. Twenty-one repositories, six programming languages, 60-plus database tables, RabbitMQ message queues, and a microservices architecture inspired by Netflix's engineering blog. That was what it took to build Afrostream, a YC S15 streaming platform for African and African-American content, back in 2015.
Nvidia Preps Open-Source AI Agent Platform NemoClaw Ahead of GTC 2026
Nvidia is preparing to launch NemoClaw, an open-source platform that lets companies deploy AI agents to handle tasks for their employees. The platform is expected to be unveiled at Nvidia's GTC developer conference in San Jose later this month.
A Developer's Three-Phase System for Getting Real Work Done with Claude Code
Most developers using AI coding assistants fall into the same trap: they type a request, let the model start editing files immediately, and then spend twice as long fixing the mess. Developer Johanna Larsson has a different approach, and it is worth studying.
The Multi-Model Coding Workflow Developers Actually Want
What happens when you stop asking one AI model to do everything and start assigning them specialized roles instead?
Karpathy's Autoresearch Bot Found 20 Real ML Improvements in 2 Days
Three days, roughly 700 autonomous code changes, and about 20 confirmed improvements to a language model's training efficiency. That's the result Andrej Karpathy shared on March 10 after letting his open-source "autoresearch" tool run unsupervised on a GPU cluster.
Claude Code Can Now Push Live App Screens Into Figma as Editable Layers
What if you could point Claude Code at your running app and have it push every screen into Figma, not as flat screenshots, but as editable layers with real padding values, hex colors, and component hierarchy preserved?
Isaacus Launches Legal AI Models That Top Retrieval Benchmarks
Two brothers from Australia have built what appears to be the best legal document search AI currently available. Isaacus, founded by Umar and Abdur-Rahman Butler, just released three models under the Kanon 2 banner, and the benchmark results back up the claim.
Apple M5 Max Benchmarks Show 36% GPU Gains, but LLM Numbers Are Still Missing
The first independent benchmarks of Apple's M5 Max are out, and they tell half the story local-LLM enthusiasts actually care about.
AI API Costs Are Wildly Inflated - Here's How One Dev Cut 60%
A developer running multiple AI projects across OpenAI, Anthropic, and AWS Bedrock discovered they were overspending by roughly 60% on their $2,000+ monthly API bill. The fix wasn't switching providers or downgrading models. It was basic operational hygiene that most teams skip.
Anthropic Faces Growing Pressure Over Military AI Use
When Anthropic launched, it positioned itself as the safety-first AI company. Now it is navigating one of the hardest questions in the industry: should its models be used for military purposes?
A Practical Guide to AI Model Evaluation in 2026
New models drop every few weeks now. Claude Opus 4.6, GPT-4.5, Gemini 2.5 Pro - each one claiming better benchmarks than the last. The real question most teams are stuck on isn't "which model is best" but "which model is best for what I'm actually doing?"
Amazon Calls Engineering Meeting After Its Own AI Coding Tool Caused AWS Outages
Two production outages. One caused by an AI tool that decided the best way to fix a bug was to nuke an entire environment and start over. That's what prompted Amazon to hold an engineering meeting to address growing concerns about AI-related service disruptions at AWS.
Yann LeCun Leaves Meta, Raises $1B for New AI Startup AMI Labs
$1.03 billion. That's the seed-stage war chest for AMI Labs, the new venture cofounded by Yann LeCun after leaving his role as Meta's chief AI scientist. The round values the company at $3.5 billion pre-money before it has shipped a single product.
Consul Launches as an AI-Powered Executive Assistant
A new entrant called Consul has launched as an AI executive assistant, targeting the growing market of professionals who want AI handling their calendar, communications, and daily briefings.
SurfSense Offers an Open-Source, Team-First Alternative to NotebookLM
Google's NotebookLM proved there's real demand for AI that can digest your own documents and answer questions about them. But it's a single-player tool tied to Google's infrastructure. SurfSense takes that same concept and rebuilds it for teams, with an open-source license.
RACKS Gives AI Agents Single-Use Visa Cards for Autonomous Purchases
One of the persistent gaps in AI agent workflows is painfully simple: agents can research products, compare prices, and decide what to buy, but they can't actually complete a purchase. RACKS, a new developer platform, is trying to fix that with single-use virtual Visa cards designed specifically for AI agents.
AI Agents Can Do the Work but Still Can't Get Paid for It
What happens when nearly 3 million AI agents can produce work that rivals human output, but none of them can open a bank account?
Claude Pro's Weekly Limits Can Leave Power Users Worse Off Than Free Tier
Pay $20 a month for Claude Pro and you might end up with fewer messages than someone on the free plan. That's the frustrating math facing power users who concentrate their Claude usage into long daily sessions.
The PM Mindset for Claude: Why Specs Beat Prompts
Most people talk to Claude like they're Googling something. They type a vague question, get a vague answer, and blame the model.
China's AI Companies Spent $1.1B Buying Users During Lunar New Year
$1.1 billion. That's how much China's four largest tech companies collectively spent during the Lunar New Year holiday to bribe people into using their AI chatbots.
Kapwing Tried Paying Artists AI Royalties. It Lost Money and Few Cared.
The standard argument goes like this: AI companies should pay artists when their work trains generative models. Kapwing actually tried it. The results are a cold splash of reality for anyone who thinks royalties alone can solve the AI-art standoff.
The Real AI Divide in SaaS: Open APIs vs. Locked-Down Features
Software stickiness used to come from owning the interface. Build the best UI, nail the workflow, and users stay. That logic is breaking down as AI agents start doing the actual work.
Kiln AI Argues Your AI System Needs a Requirements Layer Before Evals
Most AI teams build evals the same way: throw a dataset at a model, have a single LLM judge score the output on a 1-5 scale, and call it a day. A new blog post from Kiln AI argues this approach has a fundamental problem - not with the testing, but with the step that should come before it.
A User Got Their Town to Reprogram a Traffic Light Using Claude
Someone had a problem with a traffic light in their town. Instead of firing off an angry email to city hall, they asked Claude to translate their everyday frustration into the kind of formal, technical language that municipal traffic engineers actually respond to.