AWS Machine Learning · 2h
#
OpenFold3-preview, an open model predicting 3D structures of proteins, DNA, RNA and ligand complexes, is now deployable from Amazon SageMaker JumpStart in a customer's own AWS account, with async inference and multi-GPU scaling.
Why it matters: Self-hosting agent infrastructure lets enterprises run Claude agents inside their own environment.
Anthropic now lets customers self-host Claude Managed Agents, but with a catch: eight hours to act if the system flags a safety risk.
The Interactive Advertising Bureau, a digital advertising trade group, released a framework giving the industry a way to account for the role AI plays in influencing or completing purchases.
Fortreum introduced a Kovr-powered VDR/VER service to help cloud providers operationalize FedRAMP's new vulnerability rules before the December 7, 2026 migration deadline, combining security specialists, implementation support and AI-assisted evaluation.
Nvidia's cuPhoton is a GPU-accelerated library for black hole interferometry, tackling imaging challenges such as M87*'s 42-microarcsecond ring.
Why it matters: The Fed warning signals regulators may tighten scrutiny of AI-fueled market optimism.
The Federal Reserve's semi-annual financial stability report flags potential vulnerabilities from artificial intelligence and elevated asset valuations, with AI topping worries about the financial system.
Why it matters: Gives agent builders an open, self-hostable browser runtime instead of closed web automation stacks.
Kitesurf is an experimental browser engine built for AI agents. Cloudflare is open-sourcing it so developers can inspect the code, run it locally, or deploy their own instance on Cloudflare Workers.
Why it matters: One of the largest AI lab IPOs tests public-market appetite for frontier AI valuations.
Wall Street investors attended a J.P. Morgan fireside chat with CEO Dario Amodei as Anthropic prepares for its Nasdaq debut, with investors drawn to the IPO but questioning how durable its AI lead is.
GrubMarket, a SaaS company serving food producers, wholesalers and retailers, launched an AI agent that converts vendor invoices and receipts into purchase orders.
Why it matters: A large merchant base joining agentic shopping signals mainstream uptake of AI commerce.
Taiwan's Neweb Technologies partners with Mastercard on an AI shopping platform, with 400,000 merchants joining agentic commerce, BigGo Finance reports.
Why it matters: A standard agent protocol spanning Shopify and Muse could shape how shopping agents interoperate.
Sierra released Personal Agent Protocol 1.0, or Poppy, defining how personal AI agents interact; it works automatically with every Shopify store and Meta's Muse agent, with 100+ companies as partners.
Why it matters: Chip-export enforcement and Nvidia's systemic role are becoming central policy questions.
A Democratic senator wants an assessment of whether Nvidia is too big to fail, amid a forensic analysis linking Chinese AI models to Nvidia Blackwell systems, $11.4M in chips vanishing in shipment and Treasury frustration over chip policy.
Why it matters: Nvidia's executive turmoil could reshape the flow of capital into AI startups.
Nvidia poured tens of billions into AI startups under Jensen Huang; without him at the helm, founders and VCs brace for smaller checks and lower valuations, while names like Chris Malachowsky and Arm's Rene Haas surface as possible successors.
Alarm.com customers can now search their security system in plain English using AI, making it easier to find events and footage without manual filters.
Why it matters: Brings AI assistants into personal finance and account-data access, a template for consumer AI agents.
Google is rolling out a Google Wallet connection for all Gemini app users in the US, letting them ask personal-finance questions informed by spending and optionally linked bank and credit-card accounts.
Affirmi AppSpace renamed itself TheMVP.ai and added Reddi, an AI agent aimed at youth sports; the release gives no further product or pricing details.
LexisNexis released a detection tool aimed at identifying AI agents, according to a RELX stock report.
PixVerse launched a text-to-video tool that turns written prompts into directed AI-generated video clips.
Why it matters: Signals Anthropic pushing Claude into vertical, industrial operational systems.
Anthropic's Claude for Manufacturing plugin connects the AI directly into factory software, letting it read plant records and diagnose equipment breakdowns to automate industrial workflows.
Sino Biological's 4RISE is a high-throughput wet-lab service that validates AI-designed protein binders, expressing proteins cell-free in as little as three hours and delivering binding data in about three days.
A remio post says the Claude Fable 5.2 Higgsfield launch offers a reported three-day free trial. The post gives few specifics beyond the trial and the two product names.
Gopuff launches an ordering integration with Grok Bot, letting users place grocery orders through the assistant; Grok Bot can now also join meetings and take work from the call.
Why it matters: Puts mainstream AI agents directly into checkout, a major step for agentic commerce.
OpenAI launched ChatGPT Wallet in the US, giving ChatGPT agents a card wallet so they can shop and check out for users.
Why it matters: Turns DeepSeek's consumer app capabilities into building blocks for third-party apps.
DeepSeek expanded its developer API with search, file reading, voice and safety features, exposing the full DeepSeek app experience; OCR is priced at $0.225 per 1,000 pages.
Tokyo-based Licklider launched a free SciGround MCP beta in Claude Web that checks two-group study designs, ahead of a fuller design-to-claim chain.
Why it matters: Extends AI agents from chat into real-world telephony actions.
CALL-E enables AI agents inside ChatGPT, Muse and OpenClaw to place real phone calls on their users' behalf, according to USA Today.
Coinbase introduced Agent Controls, letting users set permissions and guardrails for AI agents that trade on their accounts.
Commvault announced expanded AI security and cyber resilience capabilities, according to Security Informed, aimed at protecting enterprises and their AI workloads.
SOFIA for Business, built with Oxford Saïd and ElevenLabs Impact, is a voice tool that lets founders talk to and add to Oxford research on how women build companies.
ElevenLabs introduced Forward Deployed Creatives in ElevenCreative: specialists who build workflows with a team or produce finished, on-brand creative for approval.
Translated released Lara Mini, a 7.2B-parameter mixture-of-experts model that activates 600M parameters per token. Trained from scratch on 6 trillion tokens across English and 44 other languages, including all 24 official EU languages.
Cardless launched Cardless AI, an AI feature for its co-branded credit card platform aimed at cardholders and partners.
Hyundai Card unveils PLCC 2.0, an updated private-label credit card model built with AI tools for card partners, adding AI-driven features for partner issuers and cardholders.
Why it matters: Control of product data decides who wins as AI agents take over shopping and checkout.
Amazon is reportedly building a plan to index and own the world's online product data, aiming to defend its shopping business against third-party AI shopping agents.
Claude Duet is Anthropic's consumer AI agent that can join video calls, take notes, draft plans and create action items in real time, aimed at non-coders.
Why it matters: A 1.0 milestone with API stability gives local AI tooling a sturdier foundation.
llama.cpp, the open-source C/C++ inference engine for running LLMs locally on CPUs and GPUs, hit v1.0.0, promising a stable API for local AI apps.
Two ChatGPT voice modes disappear on Dec. 11; existing chats and recordings stay, and the newer GPT-Live voice mode takes over.
Why it matters: Moves ElevenLabs from voice generation into the crowded AI meeting-notes and dictation market.
ElevenLabs introduced ElevenAI, an app combining meeting note-taking and cross-app dictation for teams, plus a medical transcription tool and expanded voice models and capabilities.
Xiaohongshu (RedNote) is testing a standalone app called Dada that lets users turn their expertise into AI agents others can use, covering tasks like design, job hunting and studying.
Frammer AI launched a live sports platform that turns match footage into social media clips in under 30 seconds, with Astro Malaysia as its first customer.
GENTO is a humanoid robot built from the sensor up, with design detailed at its CoRL 2026 debut. Its makers argue sensor-first architecture improves perception and control versus retrofitting sensors onto existing bodies.
Nvidia's cuBLAS adds new FP64 dense general matrix-multiply algorithms to speed up high-precision scientific simulations such as molecular chemistry and materials science.
UGREEN's HomeAgent collection includes cameras, speakers and more that bring the bulk of smart-home functionality and AI processing directly into the home, avoiding cloud subscriptions.
A Hugging Face blog tutorial shows how to build a robot pick-and-place simulation using SimReady assets, guiding developers through setting up a simulated manipulation environment.
Cornelis introduced CN6000, an 800-gigabit networking family aimed at cutting time AI accelerators and supercomputers wait for data, claiming up to 1.6 billion messages per second on its Omni-Path SuperNIC.
Why it matters: Removes a data bottleneck for humanoid robot training by reusing human motion capture.
Nvidia's SOMA Retargeter is open-source software that translates human motion capture into movements for humanoid robots, with support for five robot models.
Komprise and Spectra Logic launched tape tiering aimed at AI data storage, moving cold data to tape to cut cost.
Why it matters: Extends Apple Intelligence into home security video, tying AI features to iCloud+ subscriptions.
Apple's HomeView uses Apple Intelligence to bring AI video features to home security cameras, but only for premium iCloud+ subscribers. Apple says designing HomeView was a hard but fun journey.
Vertiv's PowerStation 3100 is a remote power panel for AI/HPC racks above 100 kW, with 600–1,200 A configurations on one platform. Available now in North America.
Why it matters: IPO filing lays bare the debt and Nvidia dependence behind neocloud buildouts.
Neocloud Nscale revealed $5.5 billion in debt, a need for more cash and greater reliance on Nvidia than previously disclosed in a filing ahead of its planned IPO.
CodeNinja Atoms published open reference architectures for deploying sovereign physical AI, with complete system designs for energy, oil and gas, ports, rail, manufacturing, heavy industry, aviation and agriculture.
Why it matters: Export-control evasion at this scale would reshape US chip policy and global AI compute.
A report says Chinese AI datacenter projects worth over $3 billion are powered by Nvidia processors despite US export controls. Washington officials trade blame over the Nvidia chip scandal; researchers say Blackwell fingerprints appear in Chinese models.
Shinsung E&G showcased integrated cooling and energy systems for AI data centers at Data Center Korea 2026, including its AIO module that combines servers and cooling equipment.
Researchers at IIT Madras created Seanav, a real-time digital twin for maritime navigation that combines AI and VR to give autonomous vessels a virtual ocean test environment.
UBTECH has begun selling its robots to automobile dealerships, per Gasgoo, expanding humanoid and service robot sales into retail automotive showrooms.
HiSilicon's CS6286 packs a 50 GOPS AI engine into an 8MP sensor; it detects people with 95%+ accuracy and lets dashcams sleep, extending sentry mode past seven days.
Why it matters: A world model built natively on Ascend chips signals Huawei's push into physical AI.
Huawei Cloud will unveil CloudRobo-WM1.2, billed as the first Ascend-native world model from its Industry AI Dream Factory, at an online launch on Nov 16; specs are not yet public.
Why it matters: A major PC maker entering humanoids with a factory-proven packing robot signals broader hardware competition.
Asus announced its first humanoid robot, Gallop, which learns to pack tiny motherboard parts at human speed in a factory. Asus also posted record Q3 revenue of NT$258.8B ($7.86B), up 36% YoY on AI server growth.
Why it matters: A city-backed general control model could accelerate humanoid deployment.
Beijing introduced a general whole-body control model aimed at moving humanoid robots toward real-world operations, according to the city's international web portal.
Pure Storage, now Everpure, announced new Portworx capabilities: Pre-staged Volumes for AI/ML workloads and general availability of other Kubernetes data-management features.
Apple says tvOS 27.2.1 brings Siri AI to second- and third-generation Apple TV 4K models, including 2021 units, alongside the new Apple TV 4K and HomeView launch.
WHOOP launched WHOOP Humans, a social network where each user has a 'verified heartbeat.' It turns sleep, recovery and workout data into a social feed and aims to replace bots with humans.
Gates launched the ThermalCool 2100SS, a stainless-steel coolant-circulation pump for rack-level AI and high-density computing that cools its own motor with the coolant it moves.
Innodisk announced an expansion of its AI data-center portfolio, pairing enterprise SSDs (3TS9-P, 4TS2-P, 5TS-P series) with DDR5 and CXL memory for cloud service providers and neocloud operators handling AI training and inference.
Why it matters: Another major consumer AI glasses push with Gemini at the core.
Samsung unveiled 41-gram AI glasses built with Google Gemini and eyewear partners, while Warby Parker bets on Google AI glasses for growth even as privacy concerns persist.
gpucanvas is a TypeScript open-source tool that puts thousands of live DOM nodes on an infinite canvas drawn by the GPU, with an editor to select and move them, shaders and transitions, and an AI agent that can read and point at them.
Cursor 3.30.14 is out on macOS, Windows and Linux. The editor lets developers write, edit and chat about their code with AI assistance.
Antigravity 2.32.0 is available on macOS, Windows and Linux. It is an agent orchestration platform for coordinating AI agents.
AWS describes how GS Caltex connects SageMaker Canvas to SageMaker Pipelines so process engineers' no-code models run in production, cutting deployment from about a week to a few hours with batch inference every 30 minutes.
datasette-cluster-map is a Datasette plugin that plots rows on an interactive clustered map. Version 0.19 works with Datasette 1.0a42 and depends on datasette-leaflet for OpenStreetMap tiles.
Why it matters: Raises hard questions about measuring developer output when AI tools write much of the code.
KoreaTechDesk examines a repository of 520,000 lines of code built largely by one experienced developer directing AI coding tools, questioning assumptions about AI productivity.
A HackerNoon piece describes embedding a 42-desk design studio workflow inside a coding agent for AI-assisted design work.
datasette-leaflet is a Datasette plugin that loads OpenStreetMap tiles. Version 0.3 sends a referrer header to comply with OpenStreetMap's tile policy.
AWS explains how Amazon Bedrock AgentCore runtime V2 snapshots and restores initialized environments so session start no longer grows with image size, and outlines four patterns to keep work off the first request.
TRL v1.18.0 adds a model-version cache salt so AsyncGRPO and AsyncDistillation student rollouts stop pairing new weights with first-version prefix KV cache from older weights.
Why it matters: Gives agent builders an open, self-hostable browser runtime instead of closed web automation stacks.
Kitesurf is an experimental browser engine built for AI agents. Cloudflare is open-sourcing it so developers can inspect the code, run it locally, or deploy their own instance on Cloudflare Workers.
Nvidia developer post explains how to run native Deep Learning Accelerator inference in Python on Jetson AGX Orin using cuDLA, for edge AI with tight power budgets.
Perplexity released iPhone app version 26.45.0 with bug fixes and improvements, and brought its Teammates AI agents feature to the Mac.
Amazon SageMaker HyperPod Data Capture records live inference inputs and outputs to S3; a walkthrough shows capturing a support router's traffic, finding gaps with Athena and fine-tuning on the data.
The New Stack examines how DeepSeek's coding agent, which favors orderly, minimal changes, is penalized by benchmarks that reward larger diffs.
AI Manus is a general-purpose AI agent system that runs various tools and operations in a sandboxed environment; a GitHub repo gathering stars in its first week.
Why it matters: Brings AI assistants into personal finance and account-data access, a template for consumer AI agents.
Google is rolling out a Google Wallet connection for all Gemini app users in the US, letting them ask personal-finance questions informed by spending and optionally linked bank and credit-card accounts.
Google Cloud expanded AI agent monitoring on Nov 13, 2026, adding customer-controlled logging, dashboards, privacy safeguards and spending limits across BigQuery and Looker for managed and custom agents.
Amp is a frontier coding agent and development environment; its CLI, installed via npm with i -g @ampcode/cli, gets minor new builds in this version bump.
Unite.AI reviews OpenAI Codex, saying it fixed every bug the reviewer threw at it, covering its use for simple tools and tedious coding tasks.
Apple now offers $49.99 AI usage packs to Creator Studio subscribers, adding extra AI capacity whose real-world value is hard to predict.
LayerShop is a free, open-source Photoshop-style editor that runs ComfyUI in the background. It now has Generative Fill: select an area, press Shift+Backspace, pick AI Gen, and Qwen Image 2.1 inpaints locally.
Why it matters: Turns DeepSeek's consumer app capabilities into building blocks for third-party apps.
DeepSeek expanded its developer API with search, file reading, voice and safety features, exposing the full DeepSeek app experience; OCR is priced at $0.225 per 1,000 pages.
Haru-4B is a 4B-parameter audio-to-audio foundation model that listens and speaks simultaneously, like a phone call, at 7.5 Hz frame rate with ~133 ms latency. Apache-2.0 weights on Hugging Face; research base model.
Transformers, the Hugging Face library for defining and running ML models across text, vision and audio, released v5.23.0, adding Inkling 2 Small, a multimodal model from Thinking Machines Lab.
An unreleased Exposé-like iPhone multitasking interface, unseen for over 15 years, has been brought back to life using AI and emulation, AppleInsider reports.
ElevenReader, an app that reads books, PDFs, docs and articles aloud in natural AI voices, updates to 1.11.42. Voice Design now shows one preview at a time with quick-change pills, plus character voice previews.
A non-programmer used Anthropic's Claude to build a 144-square-kilometer open-world game in four days, according to a report.
A quantized NVFP4 build of Edge0-145B, a 145B frontier-grade on-device text model built on Qwen3.8-Flash-Next, is available for download on Hugging Face.
Ava is a Hugging Face Space demo for talking face to face with LiquidAI's LFM3-Omni-2B through a 3D lip-synced avatar. It runs fully in the browser on WebGPU with no backend, so voice never leaves the device.
Haru-4B is a 4B-parameter audio-to-audio foundation model that listens and speaks simultaneously, like a phone call, at 7.5 Hz frame rate with ~133 ms latency. Apache-2.0 weights on Hugging Face; research base model.
A quantized NVFP4 build of Edge0-145B, a 145B frontier-grade on-device text model built on Qwen3.8-Flash-Next, is available for download on Hugging Face.
Ava is a Hugging Face Space demo for talking face to face with LiquidAI's LFM3-Omni-2B through a 3D lip-synced avatar. It runs fully in the browser on WebGPU with no backend, so voice never leaves the device.
A dataset of sensorized human data — egocentric video, dense language annotations, head and hand pose — meant to connect perception, action and intent for training physical AI. Hosted on Hugging Face under CC-BY-4.0.
Inkling-2-Small is a new open-weight language model from Thinking Machines Lab, released on Hugging Face.
An uncensored text-to-image model built by merging ChatR Labs' rank-128 Omni adapter into Qwen-Image-2.1 and shipping quantized transformer weights, for image generation and editing; it reached No.2 on Hugging Face's Qwen rankings.
Pika H3 Flex is a six-step post-training of MiniMax-H3 that generates video with audio, using two rank-128 LoRA adapters for text/image-to-video and reference-to-video. A Hugging Face Gradio Space offers the demo; code is released.
gpucanvas is a TypeScript open-source tool that puts thousands of live DOM nodes on an infinite canvas drawn by the GPU, with an editor to select and move them, shaders and transitions, and an AI agent that can read and point at them.
AI Manus is a general-purpose AI agent system that runs various tools and operations in a sandboxed environment; a GitHub repo gathering stars in its first week.
Transformers, the Hugging Face library for defining and running ML models across text, vision and audio, released v5.23.0, adding Inkling 2 Small, a multimodal model from Thinking Machines Lab.
ONNX Runtime is a cross-platform engine for running and training ML models. Version 1.32.0 adds opset 28 support, CPU/GPU hybrid MoE offloading, new QMoE kernels and faster session init.
A Python GitHub repo of three runnable AI engineering projects: traceable RAG, prompt evaluation and guarded NL2SQL analytics.
A curated GitHub list collecting papers on AI in astronomy and astrophysics. The repo crossed 100 stars in its first week.
DeepSpeed is Microsoft's open-source deep learning optimization library that makes distributed training and inference efficient. Version 0.20.0 adds changes including GPTBigCode shared key/value gradient summing and AutoEP expert storage maps.
An agent skill for Claude Code and Codex that builds a playable browser 3D action game in a single HTML file with AI-generated assets, using three.js. Includes a playable demo.
Diffusers is Hugging Face's open-source PyTorch library for running and training diffusion models for image, video and audio generation. v0.43.0 adds five modular model integrations across image generation, editing, video with audio, speech-driven animation and world simulation, plus attention and LoRA improvements.
Perplexity released iPhone app version 26.45.0 with bug fixes and improvements, and brought its Teammates AI agents feature to the Mac.
Google is rolling out a Google Wallet connection for all Gemini app users in the US, letting them ask personal-finance questions informed by spending and optionally linked bank and credit-card accounts.
Apple now offers $49.99 AI usage packs to Creator Studio subscribers, adding extra AI capacity whose real-world value is hard to predict.
LayerShop is a free, open-source Photoshop-style editor that runs ComfyUI in the background. It now has Generative Fill: select an area, press Shift+Backspace, pick AI Gen, and Qwen Image 2.1 inpaints locally.
An unreleased Exposé-like iPhone multitasking interface, unseen for over 15 years, has been brought back to life using AI and emulation, AppleInsider reports.
ElevenReader, an app that reads books, PDFs, docs and articles aloud in natural AI voices, updates to 1.11.42. Voice Design now shows one preview at a time with quick-change pills, plus character voice previews.
A non-programmer used Anthropic's Claude to build a 144-square-kilometer open-world game in four days, according to a report.
An AI enthusiast used Claude AI, wiring skill, and $15 of components to get a 1994 Jeep Cherokee to pass California smog tests.
Xiaohongshu (RedNote) is testing a standalone app called Dada that lets users turn their expertise into AI agents others can use, covering tasks like design, job hunting and studying.
GAtrends is a trend radar for chosen keywords that watches seven sources, learns each keyword's normal level and sends Telegram alerts on spikes, with a dashboard, API and Claude/ChatGPT integration.
Pi, a personal AI assistant app for iPhone and iPad, released version 26.1109.0 with bug fixes and performance improvements.
Speechify, a text-to-speech app that reads documents, articles, PDFs and email aloud, released iPhone version 5.5.20 with home screen widgets and a new tab.
Acer launched a classroom-of-the-future platform at IT Month and EdTech Taiwan, spanning AI teaching assistants, AI agents and a smart-campus management platform.
Kimi's iPhone and iPad app reaches version 3.1.9, upgrading K3.1 Pro for 3D creation, coding, writing and financial analysis and adding a 3D game maker.
Meta's AI moderation is banning thousands of innocent Facebook and Instagram accounts, with users reporting no real way to appeal.
Wispr Flow is an AI voice keyboard and speech-to-text app for iPhone and iPad. Version 2.0.1 tunes dictation for smoother, faster voice input.
Report says Samsung might add AI agent capabilities to its Bixby assistant in One UI 9.5; no confirmed release details.
A report describes older Chinese people learning AI tools, video editing and drone use, reflecting wider consumer adoption across age groups.
Yogurt is a free, open-source AI meeting-notes app that runs entirely on a Mac, keeping calls, transcripts and summaries local with no account required.
Adobe AI tools are coming to the Windows 11 Photos app; details on who gets access first remain limited.
Hacker News discussion highlights complaints that AI-generated app submissions are overwhelming Apple's App Store review process, making approval slow and erratic for synthesiser app developers.
Claris is testing Harbor, an app for running local AI across multiple Macs. Features, compatibility and release details are limited.
Raycast, a keyboard-driven launcher for Mac, Windows and Linux that controls apps and tools with a few keystrokes, ships versions 2.12.0.0 and 2.12.1.0.
Grok Bot on Mac gained integrations with Apple apps including Messages, Mail and Notes, letting the assistant work across more of the desktop.
Google is moving digital house keys out of individual smart lock apps and into Google Wallet on Android phones, per ZDNET.
Cursor 3.30.14 is out on macOS, Windows and Linux. The editor lets developers write, edit and chat about their code with AI assistance.
Antigravity 2.32.0 is available on macOS, Windows and Linux. It is an agent orchestration platform for coordinating AI agents.
AWS Machine Learning · 1h
#
AWS describes how GS Caltex connects SageMaker Canvas to SageMaker Pipelines so process engineers' no-code models run in production, cutting deployment from about a week to a few hours with batch inference every 30 minutes.
datasette-cluster-map is a Datasette plugin that plots rows on an interactive clustered map. Version 0.19 works with Datasette 1.0a42 and depends on datasette-leaflet for OpenStreetMap tiles.
KoreaTechDesk examines a repository of 520,000 lines of code built largely by one experienced developer directing AI coding tools, questioning assumptions about AI productivity.
A HackerNoon piece describes embedding a 42-desk design studio workflow inside a coding agent for AI-assisted design work.
datasette-leaflet is a Datasette plugin that loads OpenStreetMap tiles. Version 0.3 sends a referrer header to comply with OpenStreetMap's tile policy.
AWS Machine Learning · 2h
#
AWS explains how Amazon Bedrock AgentCore runtime V2 snapshots and restores initialized environments so session start no longer grows with image size, and outlines four patterns to keep work off the first request.
TRL v1.18.0 adds a model-version cache salt so AsyncGRPO and AsyncDistillation student rollouts stop pairing new weights with first-version prefix KV cache from older weights.
Kitesurf is an experimental browser engine built for AI agents. Cloudflare is open-sourcing it so developers can inspect the code, run it locally, or deploy their own instance on Cloudflare Workers.
Nvidia developer post explains how to run native Deep Learning Accelerator inference in Python on Jetson AGX Orin using cuDLA, for edge AI with tight power budgets.
AWS Machine Learning · 6h
#
Amazon SageMaker HyperPod Data Capture records live inference inputs and outputs to S3; a walkthrough shows capturing a support router's traffic, finding gaps with Athena and fine-tuning on the data.
The New Stack examines how DeepSeek's coding agent, which favors orderly, minimal changes, is penalized by benchmarks that reward larger diffs.
Google Cloud expanded AI agent monitoring on Nov 13, 2026, adding customer-controlled logging, dashboards, privacy safeguards and spending limits across BigQuery and Looker for managed and custom agents.
Amp is a frontier coding agent and development environment; its CLI, installed via npm with i -g @ampcode/cli, gets minor new builds in this version bump.
Unite.AI reviews OpenAI Codex, saying it fixed every bug the reviewer threw at it, covering its use for simple tools and tedious coding tasks.
DeepSeek expanded its developer API with search, file reading, voice and safety features, exposing the full DeepSeek app experience; OCR is priced at $0.225 per 1,000 pages.
llama.cpp, the open-source C/C++ inference engine for running LLMs locally on CPUs and GPUs, hit v1.0.0, promising a stable API for local AI apps.
Paxper lets users design cloud resources on a diagram, review every change and apply it in their own AWS account through visual infrastructure as code.
Aven App Agent lets users ask how to do a task in an app and get a walkthrough highlighting the right controls, with teams defining what the agent can guide.
Burnstate is a native Mac menu bar app for people who code with AI: it shows Claude Code, Codex, Cursor and Vercel AI Gateway usage limits, answers permission prompts from a card, tracks session tokens and cost, and logs breaks.
An Nvidia requisition describes native Windows agent runtimes on GeForce RTX, with local Nemotron inference and policy controls for files and network egress.
ArtCraft used AI to rewrite Photoshop as open-source software; CEO Brandon Thomas says the vision now extends to rebuilding parts of the internet.
Warp, a Rust-based terminal, shipped build 0.2026.11.11.08.17.stable_01 for macOS, Windows and Linux.
OpenAI Codex was used to uncover and fix a nearly 20-year-old bug in the Radeon Linux graphics driver, per Phoronix.