All Stories

Linear adds mobile review and feedback for coding sessions

VS Code July Copilot release improves agent diffs, worktrees and dictation

Addy Osmani argues agent code quality now lives in deterministic harnesses

Matt Pocock asks frameworks to ship agent skills for their footguns

Cohere joins NVIDIA Open Secure AI Alliance for open defense tooling

Clouded Judgement charts AWS, Azure and Google Cloud growth acceleration

maharshii speeds a custom attention kernel by swapping PTX division for reciprocal

CRUX paper finds frontier agents struggle with open-ended AI research

OpenAI begins beta rollout of Sign in with ChatGPT for partner sites

realmcore lists the hidden systems work behind building agent harnesses

OpenClaw adds extended-stable releases and a public maturity scorecard

Vercel releases mcp-handler 2.0 with stateless latest-spec support

Vercel Sandbox shows one sandbox running multiple isolated agent users

ChatGPT adds Side Chat context for tabs, YouTube and highlighted text

Vercel AI Gateway adds Grok Voice Think Fast 2.0 for realtime voice agents

LangChain eval-engineering skill adds simulated users and environment design

doodlestein adds AGENTS.md rules to stop agents hardcoding stock tickers

Deedy Das shares a camera-based air-gapped file transfer tool

Filesystem-Based Memory paper tests markdown folders for LLM agents

Mem0 migrates 400M agent memories from pgvector to Turbopuffer

Stripe details its Knowledge AI Platform for non-engineering teams

LangChain launches Certified Agent Engineer exam for agent lifecycle skills

Nathan Lambert releases lecture on tool use and function calling for agents

ElevenLabs Character Casting assigns voices from manuscript dialogue

fal launches Product Photo API for consistent catalog imagery

Google Earth adds Nano Banana 2 image generation for location-grounded visuals

Eyeline Labs shares ID-V2V for identity-preserving video restylization

Gemini macOS app adds Fn-key voice dictation and screen-aware actions

Matt Pocock releases Wayfinder skill for agentic project planning

Nous and Black Forest Labs open FLUX 3 Preview inside Hermes Agent

VS Code v1.131 shows subagent progress inside the Agents window

NVIDIA NOOA proposes agents as native Python objects

OpenAI opens ChatGPT for Academic Researchers to 10,000 users

MiniMax and Fireworks open optimized M3 sparse-attention kernels

Baseten launches Model Labs for closed-weight model distribution

Kernel Forge uses MCTS agents to optimize CUDA kernels in PyTorch models

Jon Durbin reports 167K tokens per second on an 8x RTX 5090 MoE run

A.X K2 launches as a 688B parameter Korean sovereign AI model

ChatGPT nears 1B weekly users while enterprise revenue share grows

Helion lands in Hugging Face Kernels with H100 attention speedups over SDPA

Kimmonismus flags tool-call bottlenecks for GPT-5.6 on Cerebras

Cohere Transcribe lands in Superwhisper for local dictation workflows

Matryoshka Agent splits long ML engineering runs into orchestrator and subagents

Jerry Liu summarizes founder lessons on agent loops and loop engineering

Figure marks manufacturing milestone with its 1,000th F.03 robot

DeepLearning.AI launches free course on building AI code review agents

Goodfire says Silico replicated three causal interpretability papers

Artificial Analysis launches AA-Doomscroll for rapid model chart browsing

Yacine releases Agent Evals 101 crash course with Florian Brand

Reach uses codex exec for repeatable isolated model workflows

HiFi-UMI trains deployable manipulation policies from robot-free data

Claude Code status shows a major outage with elevated errors across models

Andrew Carr collects STE skills and checkers for reducing agent slop

Reach shares Codex tip to email overnight agent change summaries

Google Flow Music adds Lyria 3.5, covers, iOS and lip-sync videos

Tavus PAL Maker turns prompts into configurable video agents

Surge AI releases HANDBOOK.md for long-context agent policy following

Kling MCP tutorial turns product photos into fashion e-commerce videos

Amp Labs creates dedicated Westpac AI engineering team in Sydney

Deedy Das warns vibe-coded SaaS clones still face enterprise-grade maintenance

Google Cloud becomes NOAA's primary weather supercomputing provider

Transluce proposes oversight foundation models for scalable AI supervision

OpenBMB highlights tiny-llm, a pure Rust inference engine for MiniCPM5-1B

Poolside and Eigen launch Laguna MLX challenge for Apple Silicon inference

vLLM joins Open Secure AI Alliance to harden the open inference stack

Kimi K3 license analysis flags hosting permissions and attribution thresholds

Kimi K3 KV cache analysis highlights strong cache economics versus major models

Zhihu Frontier details why Kimi K3 serving can require 64 accelerators

Nathan Lambert releases lecture on RL regularization and SFT forgetting

Perplexity adds US-hosted Kimi K3 for Pro and Max subscribers

World Labs R2S2R turns physical robot tasks into reusable simulation worlds

Reka opens WorldModelGym leaderboard for action-conditioned world models

Hugging Face Training Agents 3 teaches local open-weight agent RL

Perplexity Personal Computer arrives on Windows with Microsoft connectors

ElevenLabs ElevenAgents adds SMS, Telegram, Intercom, and Freshdesk channels

Kunchenguid shares a permissionless agent setup using Nix and secret isolation

Vercel Connect adds 100 plus preset connectors with scoped short-lived tokens

Anthropic releases CryptanalysisBench for measuring LLM cryptanalysis ability

Goodfire edits model weights to cut RL mode-collapse labels from 94 to 5 percent

Not All LLM Reasoning Is Visible paper tests filler-token hidden computation

OpenCode Go reports request delays from an upstream Cloudflare Tunnel incident

Codeberg policy bans AI-generated content from hosted projects

Grok app builder creates and publishes functional apps on unique domains

Matt Pocock ships his Claude Code skills as an installable plugin

OpenAI publishes eight coding-agent case studies in scientific computing

OpenAI and Anthropic back Pacing the Frontier letter on automated AI R&D

HumanLayer attaches logically ordered diff trees to large pull requests

Cursor Start brings cloud agents and MCP extensions to India plan

Polsia cuts agentic AI spend from $1.2M to about $100K monthly

Music-JEPA learns a piano sound world model from actions

Ramp open-sources PorTAL for cross-model LoRA adaptation

NVIDIA Molt offers a PyTorch-native framework for agentic RL

ComfyUI tutorial builds VFX mattes with SAM3, depth, and normals

Explorative Modeling preview claims efficiency gains from training-time search

Yuliangsheng DUV production thread maps China's AI chip supply chain

François Chollet analyzes external markets as reward signals for agents

Matt Shumer publishes Gauntlet Loop guide for builder and critic agents

Cohere launches North Automations for plain-language workflows

Nemotron chip-design thread connects RTL agents to AI self-improvement

Eve agents add Slack event hooks and thread session controls

Figure robots get visitor passes and door access on company campus

Goodfire finds a bouba-kiki phonetic axis inside Llama and Gemma

Bun spawn encoding bug breaks Execa workflows and gets an agent-driven fix

Pangram hardening changes Claude Code adversarial testing results

Trump executive order deadlines point to classified AI cyber benchmarks

Meta performance reviews draw engineering-culture warnings from Gergely Orosz

Atomic Chat runs local Kimi K3 crash-scene tests against frontier APIs

Imagine API 2.0 unifies image and video generation in one endpoint

OpenAI work report measures task crossover across occupations

Startup CTO churn gets tied to AI-native pressure by Gergely Orosz

Zed previews test result indicators directly in the editor gutter

Gemini team schedules Prompt Like a Pro session for real user workflows

NVIDIA-linked paper targets near speed-of-light latency in GPU collectives

Pre2Post chess paper predicts RL gains from pretraining loss and tokens

NVIDIA finds SOAP and Muon scale better than AdamW at 100M-token batches

John Schulman says open-model training and backdoor forensics are understudied

Denny Zhou traces reasoning research from symbolic programs to natural language

Google JAXBench tests autonomous TPU kernel optimization with real workloads

llama.cpp merges GLM-5.2-Vision support with a MoonViT projector

iScienceLuvr argues Kimi K3 increases DRAM demand despite attention savings

QuixiCore-CPU provides a portable reference backend for AI kernels

Haider recommends using Grok 4.5 as a cheap worker behind Sol planning

Ethan Mollick updates his AI model guide for Opus 5 and Codex Voice

ostris experiments with a pixel-space video model built on Wan 2.1 blocks

Yann LeCun clarifies how MAE, I-JEPA and SimCLR differ

Hermes Desktop shows a ChatGPT-like agent UI with file explorer support

Sakana Fugu-Ultra v1.1 adds a Claude Code-compatible interface

The Turing Post maps 11 protocols for the modern AI agent stack

Dex Horthy recommends Claude and Codex HTML mockups before building code

Kent Beck says coding with the AI genie revived his software experiments

UCLA paper finds ChatGPT made AI-exposed Upwork jobs more price-sensitive

Base44 powers a teen-built app running a 3M dollar junk removal business

Suno adds stem-to-MIDI export, Lyric Co-Writer and screenshot-to-song tools

Ethan Mollick urges Codex and Claude Code users to build stranger playable demos

AI Toolkit adds fine-tuning support for Mage-Flow and Mage-Flow Edit

Zhihu Frontier explains how Kimi K3 competed under compute scarcity

Kimi thread argues delta-rule state space models may exceed attention expressivity

François Chollet predicts frontier model launches will become continuous updates

OpenAI interview thread links coding agents to robots building datacenters

Runway Media Router selects image, video, and audio models by cost or latency

Airtap launches an AI agent reachable through iMessage and RCS

Wasmtime 47 enables WebAssembly GC and exceptions by default

Native SDK adds terminal component with PTY, truecolor, recording, and replay

Unitree G1 teardown estimates a 6,144 dollar humanoid bill of materials

Nick Dobos flags 199 ms latency in robotic telesurgery demo

LLaDA 2.2 diffusion LLM posts agentic benchmark scores against Ling 2.6

OpenAI resets Codex and ChatGPT Work limits after near-global outage

HeyGen opens HyperFrames launch video projects for forking and remixing

Plannotator ships Vim mode for terminal-style agent plan editing

Magnific tutorial combines Nano Banana and Seedance for a fantasy video workflow

Suno shares Crook Brown workflow for chopping and splicing generated samples

Corey Ganim maps an AI assessment process before building agents

pvncher says game ideas should be fun on pen and paper before agent coding

Third Mind Notes opens TestFlight beta for Cursor-style Apple Notes

SeedAudio 1.0 and SeeDance 2.0 power an audio-first Runway performance

Oracle faces $7B power-infrastructure guarantees for OpenAI data center

Kimmonismus says China is replacing imported AI chips faster than expected

Chetas uses tokenizer prompts to flag a hidden Qwen checkpoint

vLLM AFD Plugin splits Attention and FFN services for MoE serving

Arena reports Kimi K3 passing Fable 5 on Code Arena frontend votes

Chetas spots possible DeepSeek expert-mode silent routing via reasoning traces

vLLM prepares day-zero serving support for Ling-3.0-flash

Chetas shows Gemini 3.6 Flash regressing on precision vision benchmarks

Guinness Chen explains ChatGPT Voice's asymmetric speak-read interface

fal launches Happy Oyster 1.0 for streaming real-time playable worlds

Atomic Agent beats Hermes on GAIA with local 4-bit Qwen on M4 Max

Guinness Chen says Codex should automate every clickable app action

Lovable adds instant MCP support for any Lovable app

Steipete turns a Codex reviewer loop into a graph after real GPT-5.6 calls

Daniel Mac shows Codex on Mac can be controlled from iPhone voice remote

Ethan Mollick gets Codex to write BenchBenchBench, a benchmark-audit paper

Claude Code 2.1.220 ships stability and crash-fix updates