All Stories

T3 Code passes 1,000 contributors on GitHub

Zeeg describes Opus 5.5 as a step change for coding

Capy adds Gemini 4 Carbon support for long-horizon agent work

DSH publishes v0.2.2-rc.2 release candidate

OpenAI spotlights Mitchell Hashimoto building with Codex Originals

Amp explains multi-account windows and upcoming cross-workspace access

Researchers co-train generator and judge from positive examples

Teortaxes disputes claims that DeepSeek owns 50,000 Hopper GPUs

Runway demonstrates Fable 5.2 electronic-flute design workflow

ByteDance plans a Muse-like assistant as it works to close its model gap

A paper measures semantic leakage across 13 language models

TeortaxesTex entertains sub-30B models building operating systems by 2027

Amp adds a mode dial for mapping effort levels to models

Theo reports open-weight models receive under 0.5% of T3 Code prompts

AgenticRL-Bench tests agents on 22 real reinforcement-learning engineering tasks

Practitioner says Gemini 4 Carbon built an aquarium scene in six hours

Georgia residents sue over the approval process for an OpenAI data center

Grok identifies 17 malicious comments in Dingyi's moderation workflow

DSPy updates Monty Interpreter for version 3.5

Lysios outlines an independent research institution spanning eight fields

Will Depue offers $100,000 to support amateur astronomy research

Fofr recommends sharing agent outputs early on long tasks

Wes Roth starts a Gemini-powered Factorio deathmatch for language models

T3 Code supports Z.ai through configured harness, Theo says

Lysios invites researchers, patrons, and partners to join its work

Fable 5.2 scores 75 to lead AA-Robotics v0.1

FactIQ demo traces Jensen Huang's China statements across several years

OpenAI documents backend delegation patterns for GPT-Live voice agents

Plinius introduces Lysios with BT6's 56-operator red team

E-means explainer claims 10× faster clustering than k-means

Langfuse launches public beta of its AI Gateway

Report: Impersonators obtained OpenAI briefing on military AI work

Developer reports 17.66 Elo gain from GPT-6 Astra Stockfish edits

Report: Axios says Lutnick obstructed crackdown on Nvidia chip diversions

Report: US plans export licenses for Blackwell chips in seven countries

A T3 Code leaderboard compares models on real pull-request reviews

AssemblyAI adds Vet Mode for veterinary transcription

A reviewer reports AI-generated submissions flooding information-retrieval venues

A maintainer argues AI-generated PRs are weakening open-source contribution

Matthew Berman reports editing 20-minute videos with about 10 minutes of review

Gergely Orosz asks whether cloud certifications still matter to engineers

Nptacek argues that prompt reviews reveal practical AI skill

Google DeepMind releases AlphaGenome 2 for variant-impact analysis

T3 Code users describe it as a Mac-to-Linux offboarding ramp

User says Fable 5.2 feels worse than at launch

Vercel tests programming languages for agent-driven development

Amp describes an agent-centered workflow for collaborative software teams

Paolo Anzani recommends a 70-10-20 split for agentic coding

Alem Tuzlak argues GitHub needs agent-era workflows, not replacement

Rabi Guha argues models could render software without generating readable code

Meta open-sources Astryx, a UI system designed for coding-agent access

Report: DeepSeek plans roughly 2-trillion-parameter V5 on Huawei Ascend chips

Qoduck tester reports little improvement over GLM-5.3

Google shows Gemini 4 Carbon helping build a 52,000-blade CUDA meadow

Anthropic expands long-term compute agreements with Microsoft and Google

Maxime Labonne recommends a technical overview of modern LLM quantization

SemiAnalysis says Kessel uses fewer midplane layers than Kyber

A model analyst argues the private-public capability gap is widening again

Study traces 60.33% of GPU kernel library issues to software-stack causes

Fable 5.2 builds browser game demonstrated by Matt Shumer

Gemini 4 Carbon builds a Three.js mech that stomps across New York

Study attributes 60.33% of sampled GPU library issues to build and runtime

Fish Audio releases Live 3 for simultaneous listening and speaking

Paper previews a connected design canvas for teams building with agents

Aaron Levie outlines where agent swarms could automate digital work

Gemini 4 Carbon compares Node, Bun, C, and Swift browser-tool rewrites

Amp proposes cross-workspace project and thread access with viewer and collaborator roles

DeepSeek proposes persistent cloud sessions with Router and Worker agents

Slack app security alert alleges deployed Bank code diverted gift coins

Hamel Husain compares Artificial Analysis Intelligence Index weights across versions

Gemini 4 Carbon scores 99.3 on Extended NYT Connections benchmark

Report: GPT-6 Astra reportedly finds 35,000 additional vulnerabilities

Matt Shumer shares a 17-minute guide to prompting Fable 5.2

A study tests whether language models rank moral values transitively

Fable 5.2 tops several reported coding and agent benchmarks

WIRED examines the abliteration movement and its trade-offs for model safety

Zach Tratar warns that agent-written code can make complexity harder to judge

OpenClaw v2026.11.3 fixes recovery from interrupted updates

Report: Gemini 4 Carbon agents generate 6.5 million lines of Signal proofs

GeodesicLab reports NovaAtom-Lite protein prediction results on FoldBench

OpenDDE offers an open alternative for all-atom structure prediction

ACE Music v1 debuts atop Artificial Analysis’s vocal music leaderboard

Seedream 5.5 demo highlights stable iterative edits and image-to-GIF output

Fable 5.2 turns a training-run analogy into a generated musical form

ChatGPT reaches 1.2 billion weekly active users

Resend exposes account setup through more than 80 MCP tools

Anthropic revenue chart shows quarterly sales reaching $17 billion

Hone launches Engines to tie AI work to measured business outcomes

Precigenetics measures living-cell responses for AI virtual-cell datasets

Mollick tests Fermi paradox research with Fable 5.2 and Gemini Carbon

Fable 5.2 creates playable Starry Nightrider game

Theo rejects optimization that saved 16 MB but made long-thread searches 4× slower

Attached post discusses Anthropic IPO share sales, not Omarchy wallpapers

FT: Commerce reportedly stalls Megaspeed export-blacklist proposal for two months

Stripe completes OpenRouter acquisition, with product and integrations set to continue

Unverified post claims Google will announce restricted-access Krypton

Agent UI discussion asks how to attribute actions to signed-in users

Fable models a cantilevered building and iterates against deflection

T3 Code tests tab-complete suggestions for prompt choices

Scriptc uses static contracts to optimize JavaScript without changing runtime behavior

OpenBMB releases AutoRLBench for debugging failed reinforcement-learning runs

OpenAI builds Control Room to halt research inference workloads

Higgsfield publishes code-driven Blender workflow for a jet assembly film

GPT Image 2.6 Flare generates images in 30 seconds at maximum quality

Engineers demonstrate playable games built with Fable 5.2

Simon Willison builds a multi-device app to locate an intermittent beep

Vercel Token Pass adds GLM 5.3 Flash with a million-token context

Pika contributor ports H3 Flex inference to AMD GPUs

SemiAnalysis explains how HBM5 doubles bandwidth without adding pins

Meta’s Muse Wonder, Spark, and Ember models enter Google Cloud preview

Muse Ember is reported to beat comparable models on several agent tasks

Fable 5.2 leads VIALS-AA on scientific visual reasoning

Artificial Analysis ranks Fable 5.2 first on interactive 3D tasks

Treble simulates room acoustics to generate training and evaluation audio

OpenClaw helps build and run Omarchy on a Snapdragon Surface Pro

LearnOpenCV rounds up seven computer-vision papers and model releases

Mirage OCR handles plain signs but misses accents and Indic scripts

Fable writes a custom path tracer for Higgsfield’s diner demo

ARC Prize closes its 2026 paper track after more than 500 submissions

Devin Mobile runs eight cloud computers from one interface

Google AI Studio adds one-click Maps Platform setup for app prototypes

LangChain outlines a three-step eval-driven development loop for agents

Deep Agents author argues teams should delegate tasks, not judgment

LingBot-Touch tests latent prediction across heterogeneous tactile sensors

Cekura discusses adaptive simulations for voice-agent benchmarks

OpenAI says Terra is more efficient than Sol on visual agent tasks

Grok Bot changelog adds GitHub actions and enterprise data controls

Browser Use 1.0.0 replaces predefined actions with Python over raw CDP

OpenClaw reports hundreds of users running agents on one server

Browser Use reports Muse Ember scores 63.9 on browser tasks at $0.03 each

T3 Code adds command-palette MCP setup for external agent control

Jürgen Schmidhuber says physical AI needs learning from real-world interaction

Apple’s AMP1 harness beats specialist policies on robot manipulation

Joost van de Weijer shares advice for researchers entering the agent era

Researcher reports Claude treats reasoning-effort tags as privileged input

Vercel raises included build disk space to 64 GB

Anthropic finds code-ownership framing prompts broader Claude refactors

Mastra adds knowledge, harness, latency, and agent observability features

Betterleaks scans repositories and Git history for exposed secrets

Warp shows agents proposing a fix after 37 repeated workflow failures

Stephen Haney starts a benchmark for AI design ability

Nvidia lobbying draws scrutiny over remote access to restricted chips

Developer reports Astra outperforming Opus on TLA formal verification

Researchers trace irrelevant-context errors to language-model intrusion heads

ByteDance Seedream 5.5 adds pixel-preserving local image edits

IBM plans to use Mythos to find unknown flaws in customer code

Jane Street publishes a Socratic tutor skill for AI agents

GitHub improves its pull request overview

T3 Code data shows Claude Code taking a larger share of user turns

Seele03 generates editable Unity, Godot, and web game projects

Linear Chrome extension links GitHub pull requests to Linear Review

AssemblyAI shares practical tips for getting started with hardware hacking

Godzilla Nights renders animated posters entirely in Canvas 2D

Opus 5.5 Three.js demo: developer shares code for others to iterate on

Practitioner connects Blender MCP to vork through a DSH plugin

Rauch builds a WebGL2 vinyl player that simulates sound from record grooves

Two papers warn RLVR can narrow reasoning coverage while distillation transfers hidden traits

FlexGlue uses automatically mined hard negatives for image matching

Berkeley researchers combine remote VLA guidance with fast edge control

MParakhin reports agents pushing CPUs, GPUs, and cooling beyond expectations

Dair.ai’s weekly roundup highlights seven recent AI research papers

SGLang Summit features agents tuning serving configs and GPU kernels

DINO and iBOT remain difficult to train without collapse

Researchers call for detecting subtle reward hacking before failures surface

Researchers say EMA and learning-rate decay remain hard to compare

FrankenSim targets a working, thermally accurate mechanical watch simulation

AMD's Lisa Su says Meta Muse reached 10 million activations in 45 days

AI Compass catalogs 63 AI launches from the first week of November

Theo proposes Linux boxes for remote agents alongside Mac hardware

SemiAnalysis reports TileRT on AMD MI355 reached 1,330 p90 tokens per second

QuixiAI reports elastic prefill buffers cut QuixiInfer latency by 16%

Bloomberg reportedly puts DeepSeek Harness at 92 trillion daily tokens

A discussion predicts China’s foundation-model sector will consolidate

French copyright proposal faces objections over training-data disclosure

Score centering targets instability in reinforcement-learning training

Tencent researchers propose DSpine for transition-cached decoding

GPT-6.1 Sol appears to route some prompts to a newer model

Practitioners argue fine-tuning can outperform prompt optimization on narrow tasks

Practitioner reports an 11-Elo Stockfish gain from Astra agents editing weights

Developer adds agent API-access prompt to product sign-in dialogs

RLT adapts vision-language-action models with sample-efficient real-world reinforcement learning

The Turing Post maps 20 computer-use agents and their control surfaces

Ember packages agent plugins and full-stack apps in a self-hosted Bun runtime

A living map tracks 193 AI companies across pharma workflows

Physical-world RLVR rewards can improve through better lab work

Omar Sar argues teams should customize and own their agent harnesses

A developer uses a swarm of agents to scrutinize a dentist's quote

Eric Zakariasson argues agent access needs a new control model

Muse guides a first-time builder through welding and antenna construction

A researcher describes using AI to draft and discuss ICLR reviews

Meta joins frontier labs competing to use AI for scientific discovery

Vatsal Trivedy describes persistent agent swarms using billions of tokens daily

Sarah Wooders reports Codex defaults to programmatic tool calls

Nous Research community plugin directory passes 1,000 entries

OpenClaw releases v2026.11.2 patch for missing reminders

Tester says Gemini Krypton checkpoint takes twice as long and writes twice as much code

A developer uses Opus to review work from Astra and Sol in parallel

Revolut faces criticism for weeks without an outage postmortem

Nous Research warns that silent agent changes undermine reproducibility

TURTLE uses AlphaEvolve to create 735 harder tool-use tasks

Developer reports agent-built benchmark restarted Claude Code sessions after each step

JD Pressman frames deep learning as continuous program synthesis

Jev serves more than a million tokens for no charge in one usage report

T3 Code threatens bans for bots posting nonstop low-quality pull request reviews

Simon Willison hides long AI-generated PR descriptions behind GitHub details

Opus generates a Pokémon woodblock animation with browser-only JavaScript

John Whittaker combines agents with small models for bio-ML experiments

Venice adds Krea 3 for multi-reference image editing

DeepSeek Harness reports daily active users rose 545% after desktop release

Stanford CS146S assigns students open-source pull requests for agent-era work

T3 Code will ban bots that post low-quality reviews on every pull request

Cursor reportedly ended 2025 at $1 billion in annual recurring revenue

Opus reviews years of Chrome bookmarks and suggests what to delete

Robert Martin says system-level understanding matters more than language mastery

David Zhang publishes a find-state-machines skill for software projects

tsc-ts v0.1.2 learns project-specific build settings from measured runs

cc-remote runs Claude Code remotely while keeping file operations local

Factory agents rebuild a 100,000-line codebase from a distilled specification

Astra diagnoses two long-running OBS audio and video problems

Amp adds model switching and remote desktop controls to agent workflows

LeHami applies Hamiltonian mechanics and a symplectic integrator to JEPA models

Salesforce study finds expert distillation matches joint GRPO with 42% fewer FLOPs

GameFeel Benchmark highlights an early Qwen 27B model’s size-performance tradeoff

OpenRouter says Auto Code matches Opus 5.5 benchmarks at 74% lower cost

Hermes 0.23 adds full-resolution images to supported terminals

Data-center water use projections depend on growth and local capacity

Tinker turns a stalled GPU training campaign into a sub-$100 experiment

UK teams report shortages of even small GPU clusters for AI workloads

Data-center demand raises PJM capacity costs, while retail effects vary by state

Yann LeCun distinguishes JEPA method research from scaled production models

Vercel adds DeepSeek V4.1 Flash 1017 Ultrafast to AI Gateway

User reports DeepSeek V4.1 Flash running on one RTX 5090

François Chollet argues algorithm choice multiplies the gains from scaling

Researchers question RMSNorm’s large gradients near zero inputs

Researcher reports Devin CLI aliases claude-donut-eap to silicon-high

Modal-linked demo claims 100,000 Minecraft agents start in 60 seconds

Google acquires Crusoe’s Wyoming data-center site after permitting work

A visual guide maps eight AI architectures and their deployment roles

Cerebras demonstrates Kimi K3.1 generating over 2,000 tokens per second

A Codex user reports consuming up to 1.8 billion tokens daily

The /headless skill configures Macs, PCs and Linux machines for unattended operation

Harbor says weekly token volume has doubled to 200 trillion

AI 2027 tracker adds capability estimates from Anthropic internal models

Oxford tests how LLM agents coordinate with unfamiliar partners

Theo says Claude Code and Codex account for most harness traffic

Factory engineer Dexter Horthy previews Twelve Factor Factories for agentic engineering

Anthropic maps how scientists use Claude and Claude Code

SayRay displays live captions in a lightweight speech app

Agent overseers repair unreliable software jobs and log their failures

Creator apps are adding built-in MCP servers for agent workflows

Tembo adds a skill editor and Google Drive, Docs, and Sheets integrations

Safeagents supports custom rules for environment-specific agent tool checks

METR tracker shows internal model estimates ahead of its AI 2027 curve

Perplexity Teammate can monitor and manage market orders

Thoughtworks highlights Matt Pocock’s composable engineering skills for agents

A practitioner shares an MCP workflow for agent-directed YouTube browsing

MCPHallu reports agents pass only 23.2% of unreachable-goal runs

Theo says Omarchy overlaps with most planned t3os agent capabilities

Stanford study finds information disclosure changes LLM agents’ trading behavior

Ethan Mollick says LLMs can expand investigative research capacity

Researchers compare public benchmark reporting with labs' internal evaluations

Rust security links surface unsoundness in rand and symbol interning

Nick Dobos syncs agent memory and skills from one Git repository

Dex Horthy reports manually compacting coding-agent context at 200K–400K tokens

Practitioner reports Obsidian memory system for DeepSeek 1017

Anthropic tests how extra agent work improves a persistent wiki

Unverified report alleges AI lab fired engineer over agent reviews and code merges

Google Vantage study finds AI teammates raised business-plan scores to 67.6%

Vercel offers Ampersand v0.1 free on AI Gateway through November 13

Startups are recruiting open-source contributors before posting roles

Perplexity Teammate coordinates calendars, email, vendors, and daily planning

Vincent Koc uses one synced Markdown file as an agent workboard

Factory-mono reports its productivity index reached 371 from a 100 baseline

Venice adds GPT Image 2.6 Flare and Sunburst for image generation

Venice adds Kling 4.0 Flash video generation access

Gergely Orosz says engineering tutorials are shifting toward ambitious projects

Peter Steinberger criticizes Asahi's reported GPL switch after Omarchy integration

MCP Explained visualizes tool calls, Tool Search, and Code Mode

Franken Alignment proposes one-way external oversight for agent actions

Box CEO Aaron Levie describes compounding gains from coding agents

Antigravity launches interruptible voice mode for coding agents on Desktop and CLI

VS Code updates its multi-agent workspace with accessible canvases

OpenClaw describes an agent session shepherding each weekly release

Factory brings its coding agent Droid to Microsoft Teams

scriptc v0.4.9 claims 1.83× faster typed-array loops

Amp cuts macOS update downloads from 13 MB to about 150 KB with binary deltas

User demo reports Claude Duet can access Claude Code sessions

json-render/chat streams native agent cards to Slack and Teams

Nous releases Arete-27B with an 803-prompt political-alignment eval

Artificial Analysis launches VIALS-AA, a 161-task visual reasoning leaderboard

Tomo introduces Dachi 1.0 for conversational agents

TailSteer introduces trajectory-tail steering for multi-turn image editing

Tinker adds Kimi K3.1 Pro with 256K context

Unsloth says Qwen3.8-Flash-Next Dynamic V3.1 GGUFs use 50% fewer tokens

Nous releases Arete-27B to reduce Qwen's measured propaganda rate from 22% to 1%

A practitioner reports Opus delegating coding work to DeepSeek subagents

OpenCode adds Docker server support for remote clients on port 4096

OpenCode Browse lets agents control existing browser sessions

Anthropic estimates AI could cover 27–52% of researcher time

DeepMind reports two agent-assisted weaknesses in legacy 1024-bit RSA

Stanford benchmark reportedly finds 18 models lose 20–47 points with difficult teammates

Claude Code engineer says generated commands sometimes exposed account emails

Report: PSI reportedly used a non-public Anthropic model to prove the Aaronson-Ambainis conjecture

Cursor adds one-click Vercel deployment with database provisioning

Rune demos animated interfaces on macOS, Windows and Linux

Vercel’s Guillermo Rauch says proofs could cover tens of millions of code lines

A team migrates 4,500 files off styled-components in one day

Super.engineering brings coding-agent tasks and reviews to iOS

Super.engineering lets agents render interactive React artifacts in chat

Entire and Dagger bring CI next to code and agent context

GitHub launches a desktop app for parallel Copilot agent sessions

Ray Fernando shares a workflow for improving game performance with agents

Geoffrey Huntley prototypes distributed builds with Firecracker primitives

Renderlint checks agent-built interfaces across viewport sizes

Dang proposes a typed scripting language built around GraphQL schemas

Amp’s CEO uses an agent to build and test an Android app

James Long uses Opus and Carbon to repair Prepack's partial evaluator

Skullet measures page layouts to generate matching loading skeletons

Omnigent v0.21 adds SSH sessions and inline pull-request reviews

Prepack revival fixes several JavaScript partial-evaluation failures

SphereEncoder2 generates ImageNet images in one forward pass

Hugging Face releases ML Intern Slack bot for model evaluation and fine-tuning

Fastino releases five GLiDE open-weight decision models

OpenAI image generation tops 4 billion images per week

Hugging Face schedules a Local AI 301 quantization workshop

Epoch estimates Broadcom TPU shipments doubled in fiscal Q3

Marin finishes pretraining its 535B-A23B model on 18 trillion tokens

A video explains KV caching, prefill, decode, and batching in LLM serving

Nathan Lambert adds score centering to the RLHF book

Quantization-aware distillation targets Q4_0 model quality

HazyResearch proposes measuring AI value per joule

Johns Hopkins study links one-off training facts to model discoveries

Max Planck study finds reasoning increases model agreement

vLLM adds day-zero serving support for Baidu ERNIE 5.5 Flash Base

vLLM Semantic Router routes requests between fast and deep models

llm-d documents day-zero Kimi K3.1 Pro serving on TPU and GPU

SGLang ships megakernel support for Kimi K3.1 on day one

Cartesia’s H-Net learns token-like groupings from context

vLLM highlights TPU inference work behind AgentX results

AutoSFT uses specialist agents to improve fine-tuning data recipes

GlobeBench tests whether models can predict reinforcement-learning environments

RSI-Dojo turns research papers into executable post-training tasks

webAI open-sources TwIL-8B for coding, math, and instruction following

Stanford researchers find bits-per-byte comparisons remain tokenizer-sensitive

MiniCPM-V powers a local camera system that tracks cats drinking

X-Mem reports faster SGLang decoding with offloaded KV cache

Nanbeige releases a 33B open model with a 524K-token context

Ricursive launches Ripple-65B for agentic coding and reasoning

Long-horizon agent RL rollouts can take hours to generate

Ramp data ranks SpaceXAI among its fastest-growing software vendors

Together Compiler continually optimizes inference deployments

Menlo data shows open-model token growth has not lifted spend share

Researchers report refusal classifiers interrupting automated research loops

Hugging Face streams safer RL training for coding-agent harnesses

Moonshot reportedly seeks European access to Nvidia Blackwell clusters

A paper links unique training facts to language-model discovery capacity

A study finds reasoning prompts increase agreement across language models

Zyphra says ZYFA reduces FP8 attention error and softmax work

A Claude model draws a detailed pelican using Python-generated SVG

Moonshot’s reported Blackwell plans raise remote GPU capacity questions

A Framework Desktop runs a local RL LoRA training prototype

Ricursive previews dense Ripple-65B with an 80.5 TerminalBench score

vLLM and SGLang add day-zero Kimi support across accelerator platforms

Krea discusses how a small team trains its own image models

Together Compiler continually optimizes inference deployments with agents

Artificial Analysis finds endpoint settings can cut model accuracy

A new analysis models a path toward space datacenter cost parity

Astra experiments with direct neural-network weight editing

SemiAnalysis reports Muse Spark Open inference on VR hardware

BuoyantIO introduces Kinference to manage vLLM on Kubernetes

Meta scales Muse live video calls to 12 streams per GB200

Unsloth reports Strata support for Qwen-3.8-Flash-Next Dynamic v3.1 quants

SemiAnalysis compares TPUv7 with GPUs on agentic inference workloads

Zyphra reports faster GLM generation with its MI300X inference engine

VINCO beats frontier models in a small Catan evaluation

GenPeach wins 71% of blind image matchups against eight models

Scaling-law fits can fail to identify reliable extrapolations

Cursor Agent SDK bot turns X feedback into code fixes through Slack

StreamEval benchmarks 38 real-time video models across 300 interactive sessions

LangChain adds asynchronous subagents to Deep Agents

OpenClaw Multiplayer gives teams one shared agent and context

Yacine Learning hosts a workshop on robot learning and real-world RL

Zeeg previews a devtool pattern combining CLI and MCP access

Science’s Last Exam finds agents score zero on end-to-end research tasks

Delta adds shared Claude Code sessions, nested threads and remote MCP

NEO-VLA trains visual representations jointly with action policies

Visa tests 22 shopping agents across 15 purchase tasks

Junior agents merge 34 Sentry MCP metadata PRs in under six hours

Higgsfield uses Claude and MCP to assemble a 10-minute documentary

Paulius Ztins clarifies that the harness executes model tool requests

Photon connects Dify agents to messaging channels

Science’s Last Exam scores research agents on reproducible artifacts

Rime Prism adds sentence-level checks to LiveKit speech agents

A Tsinghua-Berkeley harness lifts RoboDojo scores for a fixed model

Voice Arena benchmarks speech recognition on real phone calls

Cursor documents a code-first workflow for building agents with its SDK

Paulius Ztins says adding more tools can make agents less reliable

AssemblyAI releases Blurt, an open-source dictation tool for Google Forms

Kevin Kern groups agent decisions into one prioritized inbox

Mastra lets channel agents reply only through assigned tools

A study finds agents often notice solutions but fail to use them

OpenAI makes GPT-Live outbound telephony available to developers

Tencent Hy4 ranks first among Chinese models in Agent Arena

Paulius Ztins recommends improving one agent before adding a swarm

Google adds prompt-based voice remixing to Gemini TTS

Junior seeks teams to test its multiplayer coding-agent harness

TanStack AI reaches 1.76 million weekly package downloads

Render offers agent sandboxes without a duration cap

Conductor joins Meta to work on a new agent interface

An Effect-TS workflow defines agent monitoring workstreams as infrastructure

Baseten launches Sandboxes with millisecond standby boots

Skill-diet tests skill changes across models before keeping them

A Volvo Gemini trial exposes gaps between model and car integration

Codex computer use completes a confusing expense task through the UI

Eon serves fresh Iceberg-table context to agents over MCP

Stripe lets business agents query payments data through MCP

Vercel launches Schedules in public beta for code-defined and runtime jobs

Stanford benchmark tests 18 models on 240 difficult teamwork tasks

LangChain maps LangChain, LangGraph, and Deep Agents in a three-minute overview

A Gemini agent coordinates MCP tools for conceptual turbofan design

CopilotKit explains building generative UI apps with Google A2UI

Verne Robotics releases V0.5, an open-weight model for robot actions

Superserve offers persistent VM sandboxes for long-running agents

AgentID appears as a sign-in option in a product signup flow

A developer builds a React interface by voice-prompting a terminal agent

Factory funds Wargames attack simulations for selected companies

ServerBench reports 22 verified model-serving systems from coding agents

Venice adds Ultrafast route for DeepSeek V4.1 Flash 1017

OpenAI releases GPT-6 Cyber to selected security teams

OpenAI makes GPT-6 Cyber available to select Daybreak users

OpenClaw v2026.11.1 fixes messaging refusals and Gateway settings

Hamel Husain urges teams to run private model evaluations

OpenAI models beat Anthropic models in a Blood on the Clocktower eval

Benchflow launches FrontierPhysics for long-horizon science agents

mini-swe-agent separates autonomous coding from assisted workflows

Mastra adds trace-level latency metrics for coding agents

MLpowered explains how hidden states carry information beyond the next token

Braintrust adds tracing for MCP server requests and responses

GDPx researchers warn popular AI benchmarks can grade valid work unfairly

OpenAI pledges local cost guarantees and public reporting for data centers

GPT-6 Cyber sharply reduces out-of-scope actions in an exploit eval

Polylane brings automated cloud issue remediation to GCP

OpenRouter documents MDM deployment without stored API keys

Ofir Press distinguishes autonomous benchmark agents from coding assistants

Vtrivedy recommends keeping old evals that separate model generations

Render says a security upgrade disrupted customer-managed PgBouncers

Claude launches read-only personal finance connections in US beta

OpenAI adds Private Safety Processing for API requests in customer environments

Common Crawl releases warc2zip for inspecting web archives

Goodfire reports mechanistic shaping raises honesty from 0% to 96–98%

A developer recommends a 1,000-line cap for agent-written files

ingest-bench publishes a correctness-checked Kafka-to-Iceberg test

OpenAI describes an autonomous pipeline for Linux kernel vulnerability research

Devin Review 2.0 cuts PR review costs while speeding up analysis

A fine-tuned 2B decision model reaches frontier quality at 35 ms

Hermes Agent adds remote browser access in Hearth desktop release

Google researchers show query-echo documents can poison RAG retrieval

OpenAI adds web grounding to its Decisions API

TinyFish splits web research into cited reports and fact spreadsheets

Notion can serve as synced memory and skills storage for agents

Parallel MCP 2.0 adds decision and research tools for agents

Qdrant 1.20 compresses vectors while preserving recall

Stripe makes modeled business queries available to AI agents

Exo explores agents that evolve their own harnesses over time

Zep turns verified agent traces into reusable skills

Harvey introduces a fact graph for evidence across legal matters

Pinecone introduces a compact RaBitQ vector index for Postgres

A developer links Codex compaction to recurring backup and archive clutter

A developer uses location-based reminders for persistent agents

Recursa reportedly raises DeepSeek-V4.1-Flash legal-task accuracy from 30% to 55%

Entire documents how to attach Hermes sessions to code changes

A RAG stress test finds targeted documents can sharply degrade retrieval

Folio 1.16 adds OpenRouter support and multilingual scan recognition

DeepTutor packages tutoring, research and practice in one workspace

SenseTime launches SenseNova 6.8 Flash for editable AI-generated artifacts

Daytona says its agent execution cloud reached $100M run rate

A pharma AI landscape maps startup activity across drug development

NVIDIA and Apple take contrasting approaches to employee AI spending

FrankenPiano models piano sound from coupled physical components

ChatGPT users spot video generation in the chat interface

Google Research tests smartwatch signals for passive hypertension screening

E2B reports $120 million in annualized revenue for agent computers

GitHub Copilot explores project-level task orchestration

GitHub reports rising agent-started pull requests and Copilot agent use