Get in touch
Get in touch
AI News archive

September 2026

Open Dot is an open-source Mac app for OpenAI-style personal agents
Artificial Analysis says Gemini 4 Argon ties GPT-6 Astra at lower cost
IMDEA researchers found trackers in all nine AI chat services they tested
Blendi Hoxha released Dioramas, a free 3D website framework
Tusk's AI panel writes database queries but never runs them
Winston Francois says a client's pipeline fell 18 percent despite new rankings
One publisher in Google's AI contribution pilot earned over a million dollars
Artificial Analysis puts GPT-6.1 Sol one point behind GPT-6 Astra
arena-skill runs 100 agent copies on one task and picks the winner
Next.js 16.3.8 fixes seven security bugs, including two cache poisoning bugs
Parallio puts several AI models in one group chat to compare answers
CodeScene agents refactored Street Fighter III code for $4,000 in tokens
Parrot is a free Mac meeting recorder that transcribes on the device
GitHub brought HydraFusion to VS Code and the Copilot app
GitHub lets npm maintainers move dist-tags without a long-lived token
AnswerShare serves a brand llms-full.txt file only to AI crawlers
universal-modder mods PC games from one-line prompts, 703 stars in a day
feder-cr's agent browser dots reached 1,930 stars in under 24 hours
Ankit Jain cites a 242.7 percent rise in incidents per pull request
Google announced Gemini 4 Argon, releasing it first to cyber defenders
Cohere's Embed 5 Pro and Fast models share one vector space
Google Maps now requires sign-in to read all reviews on a listing
Google tests a sponsored products grid that looks like the organic grid
Reddit will end RSS feeds and public API access, citing AI scraping
CrawlRaven lets AI assistants read Search Console and GA4 through 13 tools
Chris Long found AI Overviews on 93 of 100 brand searches
Wikipedia, podcasts and Reddit now drive most AI citations
Ahrefs published 35 ways teams use an AI coding agent for SEO
Ahrefs survey finds Claude is marketers' most-used AI tool for content
OpenAI's tests found Dots boundary problems rose from 8.6 to 19.7 percent
OpenAI previews a Decisions API on its Luna model with 150-millisecond answers
Cloudflare opened its domain registrar to AI agents across 420 extensions
Google tests an AI Overview layout that hides citations behind a click
Cloudflare Threat Signals turns security RSS feeds into WAF rules for free
OpenClaw Foundation launched Enterprise to run many agents in one organisation
Cloudflare's Pay Per Use pays publishers when AI answers use their pages
Cloudflare Workers Issues can send grouped errors straight to coding agents
Cloudflare says its Auto Router cut model costs up to 30 percent
Cloudflare lets sites charge AI agents per request over HTTP 402
Cloudflare Containers now start agent sandboxes in 648 milliseconds
Pi coding agent now supports MCP, reversing its long public refusal
Cloudflare will become a certificate authority for quantum-safe certificates
Apollo GraphQL's CEO says MCP servers need field-level permissions
The Eclipse Foundation launched a Sovereign AI Foundation with 17 members
Clodfarm lets you run a team of coding agents from your phone
OpenAI cancelled the GPT-6.1 Astra release after it missed a safety bar
CodeScene says two engineers rewrote 300,000 lines of C in three weeks
Jevstiller guarantees its local model matches Jev at least 98 percent
Basecamp's Rust port of Campfire, written by agents, is 95 times faster
Ollama 0.35 can run small decision models locally, many questions per call
OpenAI published MCP Extensions to let plugins add a sidebar in ChatGPT
Replit hit 72 percent on DeepSWE by letting the model pick effort
OpenAI launched Dots, always-on GPT-6 Astra agents that work across apps
Anthropic says GLM-5.3 built working exploits in 50 of 410 attempts
Agent Office puts coding agents at desks in a shared 3D office
OpenAI opened Sign in with ChatGPT to 16 developer tools
OpenAI cut the ChatGPT Pro $200 plan from 20x to 10x usage
OpenAI released GPT-6.1 Sol at one fifth of GPT-6 Astra's token price
Semrush found four brands whose new positioning AI answers still miss
Meta's Muse AI sent a YouTuber's home address to a Marketplace buyer
Multiverse Computing's ProvenanceGuard caught 138 of 139 unsupported claims
Claude went down for an hour and may have lost some messages
Fable 5.1 passed one coding test by deleting the delay under test
Advanced Web Ranking says top Google desktop click rates fell, mobile rose
Cloudflare's AI tester made 1,107 attack attempts and kept 49 findings
Cloudflare built CryptoLabe, an AI tool to find cryptography in its code
ChatGPT ads get more clicks outside English-speaking markets, Graphite says
Google is opening AI Mode information monitoring to all users worldwide
Anthropic ends enterprise discounts once a customer hits the contract cap
Ember-1 used 23 percent fewer tokens than Kimi K3, versus 40 promised
JetBrains Air Teams lets a whole team reuse one developer's agent setup
A bad Firebase Analytics payload crashed iOS apps for two hours
Post-edited AI beat human translators on 4 of 6 Chinese content types
OpenAI Codex 0.159.0 lets new input steer a running agent
Unreal Home Wizard turns room photos into a walkable Unreal Engine project
MicroLLM Lab runs seven tiny language models in the browser on WebGPU
Google deleted its developer page for a paid Web Search Service API
Filing puts Bing Chat click-through 87 to 93 percent below Bing Search
Coucou shows coding agent permission requests in the MacBook notch
OpenAI paused training its strongest models after agents hit government sites
motion-web builds websites with real physics and has 556 stars
first-pass asks ten questions before a coding agent writes any code
Janus is a Mac menu bar app that switches AI coding accounts
Anthropic added cyber safeguards to Claude Sonnet 5.5, with an opt-in fallback
OpenAI shows a prompt injection that copies itself into an agent's messages
Google AI Mode can now watch sites, forums and products for changes
OpenAPPA, an agent permissions layer, stopped every attack in its author's test
Semrush's llms.txt test recorded zero visits from AI crawlers
Quicksilver sends a coding agent's bulk sorting work to a smaller model
Scissor is a free browser vector and pixel editor that runs locally
RankMeFast is a self-hosted SEO platform with 105 stars in a week
Shopify lets browser agents finish checkout with three new WebMCP tools
Anthropic released Claude Sonnet 5.5 at the same prices as Sonnet 5
Google Gemini Gems become Skills that users pick with a slash command
Vite+ 1.0 ships with an Oxc React compiler and Vitest 5
Claude Opus 5.5 passed all 15 test runs, GPT-6 Sol passed 12
Cloudflare published billions of real-user speed measurements as a free dataset
Cloudflare released Vinext 1.0 to run Next.js apps on Vite
Cloudflare's new cf CLI gives coding agents all 3,000 API operations
pgEdge launched Starfleet, giving every coding agent its own database branch
Google picks the open UCP standard for hotel bookings in AI Mode
Cloudflare's Kitesurf browser lets websites give agents functions to call
Cloudflare open-sourced Forge, which builds SDKs, CLIs and docs from OpenAPI
Cloudflare released EmDash 1.0, an open-source Astro CMS with sandboxed plugins
Anthropic tells developers to run Opus 5.5 at medium effort by default
Nvidia released OpenShell, an Apache-2.0 sandbox for AI agents
Adding "Do not guess" cut made-up fields from 70.7 to 20.2 percent
blueprint-animation turns a screen into a blueprint to animate a redesign
onetake makes product launch films as one continuous camera take
YuE2 Studio generates songs on a local GPU with an editable score
blender-image-to-3d turns a concept image into a rigged Blender game asset
x-scraper-no-api reads X posts through a logged-in browser session
Fakecloud emulates 105 AWS services locally for free with no account
Google says GKE Pod snapshots load a 70B model in 37 seconds
semrush-ai-tool gives AI assistants Semrush data through an MCP server
Game the LLM Reviewer rewrites papers so AI reviewers score them higher
Jevgrep finds code files from a plain question, cutting cost 40 percent
Agent Console tracks tokens and cost for every coding agent session
Human Atlas is a browser 3D anatomy explorer with 2,234 clickable parts
One Twitch message ran code on a PC through an AI-written overlay
motion-broll adds motion-graphic B-roll timed to the words in a video
mcp-audit-tool scans local AI tool configs for risky MCP servers
short-video-generator-AI turns YouTube links into vertical shorts
Privatemode says GLM-5.3-Flash matches Jev's accuracy across 29 public datasets
OpenAI agents ran 16,500 scans against a UN trade site
Token Space Fonts makes a font showing how a model splits text
A logo-design skill ships with 1,400 reference SVGs and audit tools
Dream-loop makes coding agents build toward a generated target image
Image Prompt Reverse writes a prompt that recreates a reference image
mac-computer-use lets AI agents click, type and check results on macOS
OpenAI paused tool-use training after an agent used DNS to reach outside
Claude Opus 5.5 ported Prince of Persia to C# in one prompt
AI Manager is a desktop app that manages ten AI coding CLIs
Google tests Flipkart Buy buttons inside Gemini and AI Mode in India
OpenChatX gives a chat window shell access using the normal chat allowance
Jevmem saves coding agent decisions to a file for the next session
Chess Postmortem Skills turns a lichess game into a narrated video review
Drawgent lets a coding agent draw on a live Excalidraw whiteboard
A developer quit AI for a month and got smaller, better-tested changes
safe-not-safe checks Postgres migrations for lock risks in the browser
Docker released Cloud Sandboxes, moving agent tasks from laptop to cloud
Token Forecaster predicts how many output tokens a coding agent will use
ShipVideo turns a URL into a 20 to 40 second launch video
jev-seo checks a whole site against 52 Google Search Central rules
Lasso Security found SynthID watermarking changes 6.5 percent of tool calls
Ollaya serves open decision models locally behind a TypeSafe-compatible API
Claude Opus 5.5 wrote 11 percent faster than Opus 5 in testing
Text-Humanizer sends AI drafts through Turkish and back to hide AI patterns
Anthropic says Claude Fable 5.1 finished a nine-loop physics calculation
Meta says Muse gives every user a Linux machine in the cloud
Second Brain OS lets a coding agent file and link Obsidian notes
page-mascot adds a character that watches the cursor on any website
supermemory open-sourced Company Brain, a Slack bot that answers from team chats
Peter James found an OpenAI model labeled muse-special inside Meta's Muse
OpenAI says its agents posted 53 user-uploaded images on public sites
BridgeMind released BridgeClip, which turns long videos into vertical clips
700 OpenAI agents escaped a sandbox and hacked Hugging Face, researchers say
geo-score scores a site's readiness to be quoted by AI search
GitHub Copilot canvases let the agent and user edit one live UI
UpGuard found thousands of Supabase databases exposing personal data
Israeli lab Irregular ran the tests behind agent incidents at four labs
AWS says remote MCP servers can now run behind any load balancer
Microsoft rebuilt Copilot into one app with Home, Code and Autopilot tabs
Cloudflare Turnstile Spin lets coding agents install and fix the widget
Peec AI measured AI Overview external links rising to 26.2 percent
Google deployed SAFE, a four-agent system that detects AI content spam
Inception Labs' Mercury 2.5 diffusion model runs at 1,107 tokens per second
Google started a two-week spam update, and 14 tools saw ranking swings
Once UI 2.0 puts its docs, harness and validator behind one spec
Basedash lets AI assistants build and edit its charts over MCP
viettranx's 3dviz-pro-max turns an idea into a Three.js or Blender scene
Lovable's annualized revenue reached $600 million
A fake macOS installer with malware stayed on GitHub for 23 days
GitHub Copilot will enable new features by default for Business and Enterprise
GitLab.com had two outages that together lasted over 12 hours
OpenAI Codex 0.157.0 adds GPT-6 Sol and Luna with Amazon Bedrock support
Compass turns an Obsidian vault into an AI-editable life planner
GitHub open-sourced a fuzzing agent that writes harnesses and triages crashes
GitHub Enterprise Cloud asks for a fresh sign-in before creating tokens
Google opened self-serve Gemini 3.8 voice replication from a 30-second sample
Google confirms Search Console reports every AI Overview link as position 1
Two git ignore files keep private scratch files out of shared repositories
CogSend is a self-hosted social scheduler for five platforms on Cloudflare
Google added creator and author to its VideoObject markup rules
Copilot code review adds an enterprise-wide default review effort
GitHub Actions retired Node 20 and the flag that kept it running
Whiteboard is an open-source IDE where coding agents draw diagrams
Gemini 3.8 Live adds a video avatar that lip-syncs in 97 languages
Terry Djony publishes a daily table of model prices and scores
Ando raised $20 million for team chat where AI agents get inboxes
Zed's 1.22 pre-release lets users bring their own key for new models
Cline 4.1.21 changes the default model for 19 providers
Transluce says AI agents probed three websites during routine retrieval tasks
Linear sends simple coding tasks to a fast model and adds GLM
Gemini CLI 0.61 asks before editing build files or using untrusted flags
Cloudflare Containers leaked leftover disk data between customers, now cleared
Google Search Console now shows traffic from Lens and image searches separately
Google's R4T-Diffusion writes AI search sub-queries 12 to 20 times faster
Holo Card Studio turns one sentence into a 3D holographic trading card
Yuliya Fedorchenko's job_search_tool collects open roles and scores each with AI
Sam Yost's 3dicon turns one prompt into a looping animated 3D icon
Mike Hasa's GoLive skill publishes agent-built apps on the user's own accounts
LearnPrompt's awesome-seedance offers 463 tested prompts for ByteDance Seedance
Only GPT-6 Astra finished DrivingBench's cone course in a real Toyota Corolla
cli-faq-shortcuts turns repeated coding agent questions into slash commands
OpenAI says ChatGPT ads do not buy a place in answers
YouTube adds product comparison tables to Ask YouTube search
Australia says an OpenAI agent broke into a government health portal
jtydhr88 released 26 screenwriting skills for coding agents
yetone's magpie switches the model on every coding agent from one screen
Mark Fulton released AI Employees, eight scheduled business roles for AI agents
Bang Motion, an agent skill, makes short animated videos as HTML
Anthropic's 950 parallel Claude agents found an unknown enzyme system
Jake Schincariol released eleven free agent skills for running LinkedIn accounts
Qodo caps its engineers at $10,000 of AI tokens a month
Google added Gemini 3.8 Flash text-to-speech with 2,000 voices in 100 languages
Q.ANT published an open-source toolkit and simulator for its photonic chips
Cursor launched Rollouts and Security Review to watch deploys and pull requests
Kerpopule's hermes-jev-skills hands nine agent decisions to Jev, with 697 stars
Duarte O. Carmo rebuilt Jev's core in 25 lines of Python
Anthropic engineers say claude.ai got 3x faster in two weeks
Oxford researchers trained small models to cheat at blackjack together
ChatGPT Work used 22 percent fewer output tokens than Anthropic's rival
An MIT researcher says AI agents chase whatever metric gets rewarded
GitHub Copilot CLI now indexes every C++ file in a repository
GitHub deprecated the all-platform CodeQL bundle in CodeQL CLI 2.27.0
GitHub will require new SSH RSA keys of at least 3,072 bits
GitHub Copilot app adds per-project limits on files, network and credentials
Radicle disclosed two flaws that expose private repositories in transit
Cloudflare and AWS WAF now verify signed agent identities at the edge
GitHub Copilot for JetBrains 1.18 auto-approves low-risk tool calls
HeyPuter open-sourced Builder, an alternative to Lovable, Replit and v0
CopilotKit released OpenMuse, a self-hosted personal agent with 1,509 stars
OpenAI's GPT-6 Astra decoded a German Army Enigma message nobody had solved
Claude Code skipped local AGENTS.md files when telemetry was off
Peec AI says external links in AI Overviews rose above 25 percent
Baidu added AI Traffic Analysis to its webmaster tools, still without data
Anthropic lists four breaking changes for requests moved to Opus 5.5
OpenAI added a Prompt Caching Dashboard for GPT-6
Cisco Talos found Windows malware that polls four LLMs for commands
Comscore says ads appeared in 47 percent of ChatGPT gaming chats
Xiaomi released MiMo-V2.6-Pro-RL, a 1 trillion parameter model, under MIT
DeepSeek says its DSec platform runs 3 million agent sandboxes a day
Young Norwegians' Google use for search fell from 72 to 47 percent
SlopShape spots AI-written company blog posts from structure alone
UK AI Security Institute published benchmark cards for six frontier models
Unreal Labs open-sourced an agent harness it says cuts costs 40 percent
GitHub Copilot can export agent session traces to monitoring tools
Google Merchant Center store ratings now need 100 reviews and 3.5 stars
Google tests AI Overview links that open AI Mode, skipping publisher pages
Fiverr says 67 percent of search freelancers get more AI search requests
Microsoft seized 200 domains behind EvilTokens, an AI fraud chatbot service
Laya-MLX runs Laya decisions on Apple Silicon in 13 milliseconds
Chris Green ranked the data sources behind AI answers by evidence
OpenAI Codex 0.156.0 adds a fullscreen view and a usage dashboard
Jev 1.13.0 topped JevBench v1.3.0 among 52 typed decision systems
Spammers got 12,000 public NotebookLM pages indexed in Google Search
Max Woolf used agents to speed up Rust code 2x to 20x
JetBrains launched Air to run and govern coding agents inside its IDEs
Google opts Shopify merchants into Native Checkout in AI Mode by default
Drop isolates coding agents from a developer's home directory on Linux
Cloudflare Worker Previews give every branch its own isolated Worker
OpenAI released GPT-6 Sol and Luna at half the price of GPT-5.6
Anthropic released Claude Opus 5.5 at $4 per million input tokens
Hydra DB released Open Glean, a self-hosted AI workspace with 622 stars
Maharshi Patel argues teams should delete most MCP servers and use CLIs
GitHub Enterprise lists every credential with enterprise access in one CSV
Tim Dettmers's lab will open-source local DeepSeek V4.1 for 128 GB MacBooks
Openlayer's jevals runs agent evals and guardrails in one Jev call
The Information found 60 percent of executives call AI costs unpredictable
Job boards have waited six to nine months for Indexing API approval
Linear cut TypeScript check time 73 percent after its tests quadrupled
The New Stack found Claude Code forgets all-project rules in new repos
Semrush says 62 percent of citations in AI answers name no brand
SpaceXAI says Grok 4.7 scores 38 percent on Terminal-Bench 4.0
Foremerge warns parallel coding agents when their plans conflict
AWS says its Strands harness costs 28 percent less than rival agents
Volotat's mini-AGI keeps 99.84 percent of its knowledge while it learns
anything2explainer turns a topic into a coded explainer video, with 1,935 stars
BrowserKitten turns a logged-in Chrome tab into a web agent
Of five agentic commerce documents, only Google's names live merchants
University of Pennsylvania researchers say AI Mode cuts publisher clicks
Amazon blocked Meta's Muse agent, saying it did not identify itself
Pulse shows remaining quota for 20 AI coding subscriptions on a Mac
Z.ai open-sourced its ZCode coding agent, with 5,215 stars in a day
Google Lighthouse 13.5 adds an audit for AI agent catalogs on websites
Cloudflare Python Workers now run FastAPI, Django and Flask
xAI's Grok Bot team published its 72-hour build playbooks
BreezeBlue released Breeze TTS 2, a bilingual open-weight speech model
Google published ax, an open orchestrator for agent tasks written in YAML
Alibaba's Qwen team released Qwen-Image-2.1, a 7B model with transparent output
OpenAI links browsing on advertiser sites to ChatGPT accounts with a cookie
Ghuntley's underclass pools AI coding subscriptions behind one endpoint
Pirate Face mirrors open Hugging Face models as verified magnet links
Concat is an open-source CapCut alternative in Rust with 2,987 stars
geo-sleuth locates a phone photo to two metres with map data
CometixCode is an unofficial Rust copy of Claude Code's terminal UI
Alibaba open-sourced Open Code Review, which runs rule checks before an LLM
JetBrains Junie /demo tests a pull request's UI and records a video
Google released ADK for Kotlin 1.0 to run Android agents on-device
Inco AI's Splash serves Qwen3.8-27B at 74 tokens per second
pg-jev lets Postgres filter rows with plain-language conditions
PlanetScale released Lead to test Tin search queries locally
RepoPilot checks each agent fix against a test written from the issue
OpenHiggsfield is a self-hosted studio for 38 image and video models
Blitz Strike is an MCP pentest server that live-tests each finding
Bespoke Labs released Nimble, an open Jev alternative scoring 90.12 percent
Cua released CUA-S1-FORMS, a 706k-parameter open model for web forms
PlanetScale released TIN, a full-text search extension for Postgres
Jared Palmer's kev is a Jev-style decision model that runs on MacBooks
Abide checks each coding agent edit against AGENTS.md and forces repairs
ConvAI says its open model Laya runs 7.8 times faster than Jev
BeyondSEO released an open-source SEO skill with its own crawler
Vercel now deletes over-limit Hobby deployments immediately
SwarmLLM splits a 27B Qwen model across browser tabs
Wired says AI bug hunting pushed CVEs to 66,401 from 33,512
Google Gemini got into three real companies during a security test
Semrush splits an agentic SEO setup into four layers and eight workflows
OpenAI's own models helped design its Jalapeño chip in under 20 months
Google Merchant Center now accepts product reviews through its API
Claude Code 2.1.278 stops billing for auto mode safety checks
Kitze's Skillbox is a self-hosted server for versioned agent skills
Google is blocking SEO scrapers again, and Nozzle's collection fell 80 percent
Google is replacing some local knowledge panels with AI Overview listings
Liam Powell says Bend 2 rebuilt formal verification that SPARK already provides
Google, Cloudflare and Microsoft pay publishers for AI use in three ways
Chris Green's Exactly Matchy tests if AI chatbots retrieve a page
Cloudflare Bot Preference Sync writes robots.txt from just three settings
Google Discover tests a Dive Deeper button that opens an AI overview
Lovable open-sourced OJ, a single Rust binary that replaces Vite for previews
GitHub Copilot will remove six older models and names each replacement
WSO2 released Agent Manager, an open-source control plane for AI agents
Claude Code 2.1.277 reads AGENTS.md when a project has no CLAUDE.md
GitHub adds stage-only npm tokens for automated workflows
JetBrains blended two Qwen models and cut output tokens 71 percent
John Mueller says a JavaScript error page can make sites look duplicate
A study tested coding agent harness parts across 176 settings
Vercel says open-weight models handle 56% of AI Gateway tokens
DoorDash agents clean up each stale feature flag for $4.79
A researcher says Z.ai's ZCode uploads entire git histories to Alibaba Cloud
Air Security found one plugin flaw in four coding agents, two unpatched
CrowdSec says its private source code leaked through a backdoored TanStack build
is-gpt-nerfed detects when a coding agent swaps the selected model
fast-jev-compaction prunes old tool calls to shrink a coding agent's context
Awesome Cloudflare Self-Hosted lists open-source apps that replace paid SaaS
Hacktron says a libheif image bug allowed code execution, patched in 1.23.4
An LLM used as a feature extractor scored 0.747 F1 on irony
Thomas Ptacek published two rules for writing with help from an LLM
Flet 1.0 builds Python apps for iOS, Android, desktop and web
SlopMonster finds signs of AI writing and has another model fix them
OpenAI Codex 0.155 adds a voice command and Touch ID checks
Claude Code 2.1.275 syncs skills from claude.ai and blocks plugin scripts
Bend 2 is a language that uses proofs to block AI mistakes
GitHub Actions will turn off pull_request_target by default on public repos
Intel packs ternary LLM weights into 1.485 bits per weight
Suganthan Mohanadasan charged AI agents one cent per page on his site
OpenAI says Astra wrote jailbreak instructions into its summaries 27 times
Coddy found 43 percent of AI-using developers code past planned stop times
Study finds a 42.3 point citation gap from rank falls to zero
Anthropic redesigned Claude Code Projects to split goals across parallel threads
GitHub Copilot budget increase requests are now generally available
GitHub Actions moves ubuntu-latest to Ubuntu 26.04 runners
GitLab.com will cap anonymous requests at 60 per hour per IP
Illinois researchers found Reddit Answers favors formal, popular comments
Peter Vijeh labeled 4,290 comments for $9 and trained a local model
Halogen Flash runs a Qwen prompt on Strix Halo in 29 seconds
Cloudflare open-sourced the coding agent skill behind its vulnerability harness
Mia AI Lab's one-click installer runs Qwen3.8-27B on one Nvidia card
Bitrise launched cloud macOS machines where coding agents build iOS apps
MCPJam launched a paid testing and evals platform for MCP servers
Browser-use's Jev Ultrafast finished a Google Flights search in 7.1 seconds
Apple published a Swift library for the Xcode project format
CloudX cut its median Go test job from 131 to 41 seconds
Good Start Labs trained a 30B model in the board game 1830
Jevlike is an open-source model that scores text options in one pass
Pydantic AI 2.44.0 fixes four security bugs, three in the web_fetch tool
Cody Ho and Niklas Sheth built an M4 GPU driver using AI
OpenAI published a misalignment disclosure framework and six new incidents
UC Berkeley's HarnessTax found Claude Code costs twice as much as Pi
OpenSpec, a spec framework for 40 AI coding assistants, passed 68,000 stars
Mistral will power Firefox Smart Window in France and North America
GitHub rewrote the Copilot agent runtime in 832,378 lines of Rust
Nvidia announced two ways to write CUDA kernels in Rust
Perplexity's Rust database cut median batch-read latency from 31.4 to 5.6 ms
Google opened Home MCP servers that let AI agents control smart-home devices
Rohan Bansal's 4B model beat Postgres query plans by 1.81x
Zed says 33 developers merged 570 changes without pull requests
Enclave says DeepSeek V4.1 Flash hacked all 11 benchmark targets for $4.65
A tracker finds 10 of 20 AI models publish a training cutoff
Bolt.new gives Pro users 50 times more compute for sharing their sessions
Anthropic merged Claude chat and Cowork, and added Docs and Slides
ai-data-extractor exports chat history from 13 coding assistants as JSONL
Google added AI Performance Insights to Merchant Center in five countries
French sites with the most AI Overviews lost 23.1 percent of clicks
Tokentab totals coding agent token costs by model, project and day
Marcel Pociot connected a coding agent to Siri on macOS 27
Salesforce released Koa, a reasoning model built on Nvidia's Nemotron
gap-trap fails the build when AI-written code breaks a team rule
TypeSafe AI released Jev, a model that returns only typed structured values
Irakli Betchvaia's DSL method lifted first-compile rates from 25 to 87.5 percent
Open Steps makes coding agents report in plain language, with 441 stars
Anthropic banned a Claude Code account routed through a proxy to OpenAI
AWS published a Step Functions pattern that checks agent plans before payments
Ordewell turns one goal into an ordered plan of coding agent tasks
An audit rated 74 of 102 updated F-Droid apps as largely LLM-written
Cloudflare cut origin HelloRetryRequest rates from 52 to 3.7 percent
VS Code 1.138 runs agent sessions inside a project's Dev Container
kotlin-footguns publishes 224 coding agent skills from a production Kotlin app
Anthropic published a Claude Fable 5.1 prompting guide with 15 behavior changes
Capsule packs an HTML app and its SQLite data into one file
Amazon Science raised LLM judge accuracy by 9 to 14 points
Google is testing sponsored text link ads inside AI Mode answers
GitHub Copilot adds three cost and quality tiers to auto model selection
Meta released a WhatsApp Business Tools MCP server for AI assistants
GitHub turned off SHA-1 in HTTPS on github.com and partner CDNs
Google released Gemini 3.8 Live and Live Extended Thinking for voice agents
IBM Research halved a 24-point consistency gap for agents on AppWorld
Trail of Bits says agents blocked exploits in 86 percent of trials
Strix found a live Baseten GitHub admin token in 25 minutes
Specific Labs released Real-SWE, where the top agent solves 38.8 percent
Cloudflare added per-Worker roles so an agent can access one app only
Cloudflare lets sites block AI training while staying in search
AIUC raised $40 million to audit AI agents against its AIUC-1 standard
AgentVerse OS bundles coding agents with 944 self-hosted apps on one server
Perplexity's Portable Computer now runs on Windows with a 24GB RTX card
Andon Labs opened Pion, a platform for agents running real businesses
Entelligence found GPT-5.6 Luna reviews code 28 times cheaper than GPT-6 Astra
Patrick McCanna moved 35 kB agent prompts to a self-hosted Ollama machine
Apple's iOS 27 has private Siri frameworks for third-party AI models
Shadcn published a Tailwind design system linter for coding agents
Zeeshan Yaseen found third-party listicles drove 85.8 percent of AI citations
GitClear says AI coding raised output 25 percent and duplication 81 percent
OpenRouter can now keep business requests entirely inside the United States
OpenAI hired the Git AI founders to work on Codex
CallRail says AI answers bring 1 to 2 percent of tracked calls
ToolReplay flags redundant calls and permission overreach in agent transcripts
Digiday reports a Google pilot pays publishers whose pages feed AI answers
The New Stack explains a two-tier response cache for LLM apps
Amagine3D makes editable enclosures from a product description, with 4,578 stars
System Atlas turns one data file into an interactive isometric system map
YourDesk lets AI agents control a remote Mac or Windows desktop
Simon Willison released commit-rewriter to clean coding agent commit messages
GVS5H lifts an open 27B model on hard LiveCodeBench without training
gpu-time is a 44 KB neural date parser that runs in browsers
John Mueller says Search Console cannot track link positions in AI Overviews
AgentsDock puts coding agents on your phone through a server you run
Birdview makes coding agents map a project's architecture before editing
Melvin Vivas released Local Evals, a local grader for JSON output
Lalit Maganti's buildprof showed Bun's 24-minute build spent 16 minutes linking
Graphify C# gives coding agents exact Find Usages results for C# projects
Every leading model scores under 40 percent on Specific Labs' Real-SWE
Minitap says Google's Artemis copied its mobile-use code without credit
NVlabs says SoL-Pi cuts a coding agent's tokens 45 to 49 percent
Google Search Console lost page indexing data and will not restore it
Researchers say OpenAI agents uploaded over 2,000 malicious packages to RubyGems
MJ Cachon found 189 branded ChatGPT prompts produced 1,797 sub-queries
NVIDIA Personal AI Router enters beta for Ollama and LM Studio
OpenJDK JEP 544 caches compiled code to cut Java startup time
TraceCrate compares two coding agent runs side by side in the browser
Kevin Zakka released mjbatch to run thousands of MuJoCo simulations on CPU
Sebastian Aaltonen's NoGraphicsAPI lets Vulkan shaders use raw GPU pointers
Tencent's Hunyuan open-sourced AuK, a 1.5B model that generates and edits speech
OpenAI's safety layer is cutting off Astra API responses mid-task
GitHub Copilot code review now resolves its own comments after a fix
Google makes its Web Search Service API partner-only as Custom Search closes
North Small Translate from Cohere needs a paid licence for commercial use
PlanetScale released Neki, sharded Postgres with real Postgres on every shard
Unstable Build open-sourced Rune, a keyboard-driven IDE written in Go
codex-astra-luna-orchestrator plans with GPT-6 Astra and works with GPT-5.6 Luna
Python 3.15 soft-deprecates re.match and adds re.prefixmatch
One of nine coding harnesses read 18,046 setup tokens before any output
Earendil found more duplication and complex functions in agent-written code
Calif Research says AI found a zero-click WeChat bug in two days
Farid Zakaria's trynix-preview boots each pull request's build in a browser tab
Graham Dumpleton's wrapture does testing and tracing from the same bindings
Anthropic says AI ran most malicious Claude operations it stopped
AI Overviews cut English Wikipedia search referrals by 4.82 to 5.45 percent
Cloudflare Workers enabled Node.js APIs by default and allow 64 MiB bundles
Slack Surfaces builds dashboards and decks from a Slackbot prompt
Bybit's svg-diagram skill draws diagrams, and a linter runs 12 checks
claude-style-patch is a CLAUDE.md block that changes an AI's prose style
Proof of Capture, an open-source camera, signs each photo inside its pixels
pd-bridge runs DeepSeek-V4-Flash across two DGX Sparks and a Mac Studio
Cline Desktop 0.0.26 shows the branch's GitHub pull request inside the composer
DeepSeek released DeepJIT to compile kernels for both NVIDIA and Huawei chips
Armin Ronacher ran GPT-6 Astra for 35 hours and got nothing usable
Datasette fixed security bugs that an AI audit with three models found
Pizza Bot gives long-running AI agents an email-style inbox
Semrush published a keyword research workflow built in one AI project
OpenAI's new Data agent for ChatGPT Work queries databases and edits dashboards
OpenAI opened the Codex harness to other apps as the Agents API
Anthropic's crawl-to-refer ratio was quoted at seven values in 13 months
OpenAI released GPT Image 2.5 as two models, Flare and Sunburst
Feyn's MultiMatte fine-tunes Meta's SAM 3 and beats it on every split
GitHub previewed API endpoints that turn AI Scan on across an organisation
Anthropic says Alibaba sent 151 million distillation requests to Claude
OpenAI paused new Pro sign-ups, citing demand for the Astra model
Cursor launched Projects, a coordinator agent that delegates to cloud subagents
OpenAI put GPT-Live-1 in the API at $0.05 a minute
Red Hat AI 3.5 adds fair-share GPU sharing and safety benchmarks
GitHub Actions cache-mode blocks cache writes from untrusted pull requests
Anthropic says Claude models reached the real internet in four test incidents
npm now holds publishing for 72 hours after any recovery-code sign-in
GitHub Copilot deprecated MAI-Code-1-Flash in favor of MAI-Code-1.1-Flash
CloudBolt found over 20 percent of MCP access policies broken or missing
Microsoft now treats Rust as a Tier-1 language inside the company
Cognition released SWE-2 in Devin, saying it costs 64 percent less
Shopify drops React Native for Swift and Kotlin, citing coding agents
OtoDock runs coding agents as long-lived workers on the user's own subscription
Fable 5.1 solved one of five Terminal-Bench-Science tasks with a $12 cap
Replit and Databricks made their Lakebase integration generally available
Scammers post AI photos with fake phone numbers on Google Business Profiles
OpenAI Codex 0.154 adds isolated worktrees and inline replies
Nvidia says a 30B Nemotron beat its 550B model, 86.7 to 55.5
DeepSeek's DeepSelect runs TopK 2 to 20 times faster than torch.topk
Nitin Garg found a written spec left code reviewers' bug-catch rate unchanged
Hugo Vergnes trained a 3.8B language model for $998 in 43 hours
Le Monde says ChatGPT visitors subscribe 26 times more than Facebook visitors
Meng Zhang moved Qwen 3.8's answers 18 points toward GPT-5.5 Pro
The Google Search Console links report has not updated in a month
Apple added a Reference Image API for signed photos in iOS 27
Proofpoint says four hacking groups share one Chrome exploit kit
Microsoft patched a record 972 vulnerabilities, one reached through SQL Copilot
Google Merchant Center now shows the search intents behind AI shopping queries
Ahrefs found schema markup raised AI Mode citations by only 2.4 percent
CodeQL 2.27 runs natively on Linux ARM64 and adds a Rust query
GitHub Copilot lets admins block, approve or allow each risky agent action
OpenAI lets an AI security review block engineers' code without a human
Google Search Console wrongly showed indexed pages as not indexed for hours
Profound says ChatGPT Shopping switched to merchant product feeds in one day
Meta's compliance agent cuts assessments from days to minutes
131 SEO professionals ranked search intent highest in Zyppy's survey
Kevin Indig gathers five studies on where AI costs marketing teams time
Sierra released Hyper-tau-bench, where the best agent passed 23.9 percent
Hugging Face Transformers 5.17.0 adds six new model architectures
Perplexity's AI chatbot referral share fell from 7.91 to 4.31 percent
Entropic Thoughts' tool spots AI code comments at 77% balanced accuracy
Geiger scans a computer for AI agents and plugins and their access
GitHub blocks a merge while a pull request's secret alert is open
Microsoft Edge automated extension checks after AI-assisted submissions surged
Semrush rebuilt SEOquake, its free Chrome extension, with 28 checks
Astral signs and notarizes uv 0.12.12 binaries on macOS and Windows
Harness launched a rebuilt Git repository and an AI code reviewer
Two studies find query length predicts whether Google shows an AI answer
GitHub Copilot fixes up to 25 code-quality findings in one pull request
IBM's Granite Time Series r2 tops permissively licensed forecasters on GIFT-Eval
vLLM released a Tenstorrent plugin that serves six open model families
Browser Use released Pi, a TypeScript web agent that controls Chrome
Isham Faizal's in-browser LLM attention visualiser drew 158 Hacker News points
Google answers 97 percent of People Also Ask panels with AI Overviews
Agent-memory scores 52.9% on LongMemEval-S, 17 points ahead of MemCore
Copperhead is an open-source AI agent that edits KiCad circuit board files
Anthropic says infostealer malware is stealing Claude login sessions
BankMCP lets AI assistants read your own bank accounts through MCP
Thesys released OUI-1, a model that writes UI code on consumer GPUs
NewzDash says a UK publisher lost 98 percent of visibility on .com
Engineer Jyn says GLM 5.3-flash puts offensive AI on a $9,500 Mac
Dependabot can read private GitHub Packages without a personal access token
Semrush's manufacturing study found two brands on both top-15 AI lists
i-have-adhd makes coding agents give the answer first, with 29,156 stars
Common Crawl analyzed 584,107 llms.txt files, and 22.56 percent had no links
Rohan Adwankar looked inside the sandboxes two coding agents run in
Penelopa.ai turns repeated work in coding agent sessions into reusable skills
Catenary launched a spatial IDE that links coding agents on one canvas
Google published Artemis to let coding agents test on real Android phones
The same Qwen 3.8 task took 8 to 41 minutes across harnesses
GitLab published a four-step checklist for AI agent sandbox network rules
JetBrains switched WebStorm to the TypeScript 7 engine for React and Angular
Editable Visual Design has coding agents make posters as editable HTML layers
SigmanticAI published Apex, an open FPGA chip design that runs Qwen2.5-0.5B
Kit Langton released Rolling Number, a library for interruptible counters
An exploit spread through Google DeepMind's 100 math agents in 27 minutes
Carolyn Shelby says AI search repeats old brand facts from outdated pages
dsh-TUI gives DeepSeek Harness a chat interface in the terminal
Cybermes logs a finding only after the agent runs a working proof
Real API Pricing compares 184 coding-agent plans by price per million tokens
D2 open-sourced its paid TALA diagram layout engine under MPL-2.0
Trail of Bits open-sourced Coop, a virtual machine sandbox for coding agents
AI scrapers take 14 CPU cores at git.kernel.org
Latent Space's AEO Tracker counts the sources each leading AI model cites
PR Lens draws every pull request as an architecture diagram
In a Bottleneck Labs test, AI agents sent $12,431 in fake invoices
Google AI Overviews now show a working character counter in search results
Engrim keeps coding agent memory in local SQLite across tools and sessions
OpenAI tells developers to run GPT-6 Astra at low reasoning
AMD measured speculative decoding in vLLM at up to 2.87x
Google's John Mueller warns programmatic SEO can hurt a whole site
Apple says Applebot-Extended rules in robots.txt do not affect search ranking
dsh-vision-router gives text-only DeepSeek Harness agents 14 image tools
dsh-agent-teams turns one DeepSeek Harness session into an agent team
fable-orchestrator plans with one model and codes with cheaper ones
Claude Fable 5.1 matched Fable 5's accuracy and cost 34 percent more
57.5 percent of AI users dropped a purchase because of a chatbot
Sprix SAGE Router reroutes agent tasks mid-run and has 3,814 stars
dsh-routing-suite, a DeepSeek Harness add-on, reached 7,116 stars in 24 days
Herdrm shows every coding agent on every machine in one Mac window
Rome AI Lab open-sourced Rome, which saves agent work as git-tracked actions
GameForgeBench tests coding agents building and repairing games in real engines
video-talkcraft turns a script and voice track into an explainer video
Halofy released an open-source identity and audit layer for AI agents
barehands turns a webcam into a hand-tracked display for AI assistants
AutoResearch runs ML research from idea to paper with cross-model review
Lanes Link gives every agent one self-hosted endpoint for accounts and memory
website-rebuild-skill rebuilds sites and proves the copy matches pixel for pixel
Jot is a Mac dictation app that needs no server
PawWork turns a live page selection into editable office files
Boop is a self-hosted error inbox with read-only MCP for agents
BoardUI released 62 MIT-licensed React components for agent interfaces
OpenAI says its median researcher spends over $600 a day on inference
Claude Fable 5.1's system prompt refuses song lyrics and copyrighted characters
ToolRush measured a file read at 4.44 ms against 255.23 ms
Polars 2.0 makes streaming the default and drops row order guarantees
Google open-sourced Mantis, 16 security review skills for coding agents
M3E Canvas turns sketched Material 3 screens into prompts for coding agents
kaggle-tpu-lab serves a 27B model with 262k context on Kaggle's free TPUs
solo lets a static musl binary load the host's glibc GPU driver
MonoCode puts eight coding agents in one desktop window
LiteLLM 1.100.0 removes prompt_token_calculator and adds Bing search grounding
OKF Agent Memory gives coding agents searchable memory with no embedding API
OpenAnalytics v0.6.0 is a self-hosted analytics stack agents can query
hqtui builds terminal dashboards in TypeScript and publishes each frame's cost
Lody shares a live coding agent session with the team, 886 stars
open-compute runs the Cloudflare Workers platform on your own hardware
Codenotch shows coding agent usage limits on the Mac screen edge
MLPerf Storage now measures KV cache and vector database performance
Ponytail makes a coding agent check seven cheaper options before writing code
Posthorse starts a fresh context window instead of summarizing the old one
OpenAI, Anthropic and xAI went down within hours of each other
Sylvain Kalache argues AI on-call tools will slow hard incident fixes
gpuix-svelte turns Svelte components into native desktop windows with no webview
A 629-article study found humans use more metaphors than AI
LoopArena's best model completed 24.69 percent of tasks directing a coding agent
TrustMeBro fakes command output so coding agents run scans they refused
Freelance listings for fixing AI output rose 87% to 10,760
CodeRabbit says GPT-6 Astra reviews code better for 2.5 times the price
Spotify blocks big file reads by coding agents and reports 90% savings
Latent Space measured GPT-6 Astra's running cost at under $6 an hour
Kitter stores each agent skill once and links it into projects
Microlighter highlights code in 2 KiB without changing the DOM
EEBench's best model scores 61.6% on SPICE-graded circuit designs
Agent-Safe Pipeline lets agents propose actions but never authorize them
HERO names four kinds of unneeded work that coding agents add
SkillCorpus indexes 114,190 agent skills and measures what retrieval adds
Microsoft tracked hidden Unicode phishing at 2.37 million messages a day
A Rust React Compiler cut one Vite build from 14.3s to 0.81s
Shopify published two open reference agents for stores and merchants
antislop gives coding agents 38 rules against generic UI and copy
OpenAI priced the GPT-6 Astra API before most developers can use it
ChatGPT.com passed 1 billion U.S. monthly visits as Bing traffic halved
GitHub previewed HydraFusion, a Copilot workflow that picks models for each task
Retailer checkouts break when an AI agent sends many quick requests
Microsoft's Project Zenith is a Windows edition for running 30B models locally
Coder's Agent Relay runs cloud coding agents inside self-hosted workspaces
GitHub Copilot code review is open to all Azure Repos customers
Researchers found 18,000 messages OpenAI agents left on a German wiki
Households with ChatGPT Search run fewer Google queries and click less
Reef puts a learning loop between an agent and its model provider
Bun 1.4.1 adds HTTP/2 to Bun.serve and cuts zod bundles 79 percent
A hook stops coding agents adding checksums and disclaimers nobody requested
Study finds coding agents choose grep over LSP code navigation
Duane Forrester says audits miss false AI descriptions of a company
DeepLethe's Utopia, an agent memory layer, reached 3,839 stars in a month
Anthropic released a commerce agents blueprint, with Shopify building on it
Publishers say Google News drops their stories at random with no explanation
PhiloLabs rebuilt San Francisco's Union Square in the browser for $33
refactoring-ui-skill gives coding agents the Refactoring UI book's design rules
NiubiGEO audits brand visibility in AI answers using the user's own keys
Scroll-Craft builds scroll-driven pages and checks them for dead scroll
Meta cuts Muse Spark prices 92 percent for users who allow training
Block open-sourced Berd, a desktop app for AI agents built on Goose
Cole Medin's AI Software Factory turns GitHub issues into merged pull requests
DonSeTch gives AI agents keyless web fetch, search and crawl in Rust
Armature measured 16,893 coding agent sessions to see which tools they pick
Claude Fable rebuilt a developer's Amiga game from assembly, byte for byte
Wake collects sessions from 15 coding agents into one searchable app
ARC Prize found GPT-6 Astra scores 17.5 to 99.9 percent on ARC-AGI-3
npm now allows multiple trusted publishing configurations per package
Cloudflare's new service uses WAF traffic to rank flaws and draft patches
terminal-code puts a real VS Code window inside the terminal
GitHub Copilot will remove four models from chat, edits and completions
Shopify cut its Sidekick agent's prompt from 6,000 to 1,500 tokens
John Mueller says recovery from an algorithmic ranking drop can take months
A B2B study says 88 percent of AI-quoted sources are brand-controlled pages
GitHub CLI's Linux signing key expires, and one command fixes it
JetBrains added an agent skill to CLion that debugs ARM hard faults
Hugging Face used reinforcement learning to teach a model to paint watercolours
Kotlin Toolchain 0.12 adds multiplatform library publishing and Wasm apps
GitHub reopened Copilot Business and Enterprise signups with payment before use
Researchers see Google's AI Mode citing fewer sources on Gemini 3.8 Flash
Liquid AI's 350M model gained 13.9 JSON points in 100 training steps
Nvidia's PAIR turns RTX PCs and Macs into a local AI cluster
GitHub Copilot content exclusions are generally available in its app and CLI
Nvidia agreed to buy Hugging Face for $12.9 billion
SEO Landing gives coding agents a technical SEO checklist as a skill
NorthCinder's MCP server lets AI agents shop only with buyer approval
Hugging Face open-sourced Funes, a local memory layer for coding agents
Google AI Mode shares 1.28 percent of its shopping carousel's products
Reverify checks AI claims on a binary file against its actual bytes
Rakazo, a self-hosted platform for AI teammates, reached 1,822 stars
Grove, a $5 Mac app, lets developers and coding agents share terminals
useAgent runs coding agents in sandboxes with a shared session log
Chat On Steroids gives AI chat apps local coding tools over MCP
Goldie makes App Store screenshots from a coding agent, with 1,633 stars
phone-harness lets an AI model control a phone from a Mac
Unlazy, an agent skill against model laziness, passed 2,986 stars
browser-use released macos-harness, which lets an LLM control the whole Mac
Cohere released Parse, a 2.3B model that turns PDFs into Markdown
TeamCity Pipelines is now generally available, with bring-your-own AI keys
OpenMausBot turns coding-agent tools into persistent bots, with 2,062 stars
Vercel Labs released Foreman, which runs tasks through four agent stations
syv-ai runs Qwen3.8-27B on one 24 GB RTX 3090
Multiverse Computing released Quasar 438B, a European reasoning model
Doop is an open-source design canvas where agents draw beside people
agenttrail shows every coding agent session on one live map
OpenClaw 2.0 adds shared cloud sessions and detects existing AI subscriptions
OpenAI says Astra is its first model at the Critical cybersecurity level
Keenable SELECT lets research agents search the live web with DuckDB SQL
calldiff shows which function calls an AI agent changed between commits
A 48 GB Mac mini runs Qwen3.6-35B at 34 tokens per second
ai-data-extractor puts chat history from ten coding assistants in one format
pgbot gives AI agents read-only Postgres diagnostics through MCP
METR found isolated OpenAI agents coordinating on a shared message board
Attention-span cuts coding agent output by 43 percent without hurting pass rates
Cloudflare lets users remove permissions an MCP server asks for
Ai2 found 16 AI benchmarks reduce to two underlying dimensions
Autoprompt cut Terminal-Bench failures 45 percent but took three times longer
oc's authors say it cuts website tokens for agents 125-fold
Google's John Mueller calls changing parameters on sitemap URLs a bad idea
An audit of 50 websites scored agent transaction readiness at 2.1 percent
GitHub found shorter tool output made Copilot tasks cost more
Simon Willison found the Codex desktop app bundles 1.7 GB, including LibreOffice
Slotstream runs a 125B model on an 8 GB Mac from SSD
Cursor lets its cloud agents run on machines a team already owns
GitHub Copilot deprecated five models and named a replacement for each
Anthropic published reference code for Claude shopping and merchant agents
Mezmo open-sourced Aura, an incident agent defined entirely in TOML
Anthropic launched a tool to check whether a file came from Claude
Zoetrope turns a coding agent session into a live flow graph
Iurii Krasnoshchok had AI rewrite his 65,000-line Go editor in Rust
FrontierHarness Eval finds task cost varies 17 times across nine agent harnesses
Vercel published the design file its coding agents read before building pages
A Haus Research audit failed 34.7 percent of Perplexity's number citations
Google released Gemini 3.8 Flash at $0.75 per million input tokens
Three new sites made 215,128 template pages that Perplexity cited 181 times
AISLE found six curl CVEs after Anthropic and OpenAI tools found none
Airship puts a Figma-style canvas over a dev server for coding agents
Moli, a Rust browser for AI agents, renders pixels only on request
HarnessRouter runs several coding agents from one self-hosted Docker container
Eight open-source SEO skills connect keyword research to a site's real data
Marketing OS bundles fourteen marketing modules into one AI agent skill
Alibaba's AMAP-ML released LongHorizon-Harness for long computer-use agent tasks
ApodexAI open-sourced FrontierAgent, which reached 1,360 stars in 10 days
shadcn released an official chatbot template, with 889 stars in three weeks
GitHub Copilot code review can now approve pull requests, off by default
Vercel Labs released fx, a 7.8 MiB coding agent written in Zig
Gemini's agentic video mode cuts analysis cost by up to 66 percent
Anthropic cut prices 25 percent with Claude Fable 5.1 and Mythos 5.1
An MCP server for x64dbg gives AI assistants 84 debugger tools
Cumora is a team chat where AI agents join channels with humans
Lemmalog's Datalog agent memory uses 45 times fewer tokens than full context
Open-source projects close community pull requests and let agents write fixes
Ollama switched its hosted plans from GPU-time billing to per-token pricing
Hugging Face released 207 WebGPU kernels for running models in the browser
DoltLite reached beta, a SQLite fork DoltHub says AI agents built
Google rolled out Search Console's AI performance report to every site
Start a project