GoodAir TechTechnology and innovation
English
EspañolCLPEnglishUSD日本語JPY中文CNYDeutschEUR
Let’s talk
/ NOTEBOOK 04 · CURRENT AFFAIRS

AI also moves beyond the lab.

We track models and technical advances, but also data governance, agent governance, security, investment, and economic decisions. We look at the United States, China, Europe, and the rest of the world with the original source always at hand.

Earth seen from a space station
01 / Global viewPhoto: SpaceX · Pexels · CC0
Infrastructure

How to create a Kubernetes runtime validated for NVIDIA Dynamo

NVIDIA Developer details how to build a validated Kubernetes runtime environment for NVIDIA Dynamo. The proposal aims to offer AI teams a reproducible path from model setup to an inference endpoint. NVIDIA Dynamo is presented as a distributed inference solution.

Open source: NVIDIA Developer
Enterprise application

End-to-end management of Amazon Bedrock costs: playbook for FinOps teams and platform

The article presents a five-stage operating model for managing costs in Amazon Bedrock: assign every dollar to an owner, make all requests observable, enforce budgets, optimize platform defaults, and sustain a review cadence. Aimed at FinOps and platform engineering teams.

AWS Machine Learning
Enterprise application

Preview live web apps while your code agent builds them with the Amazon Bedrock AgentCore runtime

Amazon Bedrock AgentCore session endpoints assign each session an authenticated URL that forwards any HTTP method, routes, and WebSocket connections to allowed ports. This lets you run a code agent that keeps a Vite site in the session and see it update in the browser as you edit files.

AWS Machine Learning
Infrastructure

Run AI models to build apps at the edge with few commands

According to NVIDIA Developer, AI is moving beyond the data center: developers are creating edge agents that, among other uses, monitor driver attention and cabin conditions. The proposal lets you spin up models with a few commands to streamline edge application development.

NVIDIA Developer
Enterprise application

Worldwide Express automates LTL freight exceptions with a multi-agent solution on Amazon Bedrock and reduces active cases by 75%

Worldwide Express automated its LTL freight billing exceptions process with a multi-agent solution based on Amazon Bedrock. In six months, it reduced active exceptions by 75%, shortened time-to-cash by four weeks, and unlocked US$10 million in free cash flow.

AWS Machine Learning
Infrastructure

NVIDIA NemoClaw: build specialized agents on a stack you control

It explains how to build autonomous agents with a model, an orchestration framework, and domain-specific tools. For safer deployment, developers need a runtime that enforces controls. NemoClaw is oriented to enabling these agents on a stack under the team’s control.

NVIDIA Developer
Research

From synthetic data to real trust: evaluating Microsoft Copilot at enterprise scale, per Microsoft Research

Microsoft Research examines how to evaluate Microsoft Copilot at enterprise scale without compromising privacy. Validation requires test data that reproduces the complexity of organizations, but real data contains sensitive information that cannot be used. The approach aims to close the gap between evaluations with synthetic data and operational trust in Copilot.

Microsoft Research
Models and products

OpenAI will make it easier for websites to express AI training preferences via a shared signal in robots.txt

OpenAI will honor a unified signal in robots.txt that allows website operators to communicate to participating providers their preferences about using their content for AI training.

OpenAI
Models and products

OpenAI community safety report and new protections for teens

OpenAI bolsters violence prevention in ChatGPT with expanded safeguards, earlier interventions, and a new commitment to public safety reporting, in line with its community safety report and teen‑focused measures.

OpenAI
Models and products

OpenAI shares excerpts from court filings on how fair use supports AI training and innovation

OpenAI releases excerpts from briefs filed by creators, academics, accessibility advocates, and legal experts, asserting the role of fair use in supporting AI system training and promoting innovation.

OpenAI
Open ecosystem

From database scoring to RL rewards: post‑training agents with ThinkingBox

An approach is presented that turns grading in databases into reinforcement learning (RL) rewards to tune agents after training, with ThinkingBox. The goal is to link structured evaluation with agent improvement through rewards derived from that grading.

Hugging Face
Politics and economy

The EU holds a high-level dialogue on SMEs in competitive, resilient value chains; chaired by Stéphane Séjourné on November 17, 2026

On November 17, 2026, Executive Vice-President Stéphane Séjourné will chair a high-level dialogue to assess how EU policies, such as the European Competitiveness fund, can strengthen SME participation and scaling in critical value chains and address pending challenges.

European Commission · Internal Market
/ VISUAL ARCHIVE · OPEN SCIENCE

Scientific observation, data, and instruments open new questions.

Bright magnetic loops over the Sun’s surface
Solar magnetic fieldsNASA Images
Two satellite views of Hurricane Juliette and the height of its clouds
Atmosphere turned into dataNASA Images
Cosmic Cliffs of the Carina Nebula observed by the Webb telescope
Structures on a cosmic scaleNASA Images
Scientific map of height anomalies in the Gulf Stream
Ocean currents and heightNASA Images
/ THREE REGIONS TO WATCH CLOSELY

The same progress looks different depending on the place.

The United States concentrates a large share of leading labs and platforms. China connects models, industry, and public policy; Europe combines adoption, infrastructure, and risk-based regulation.

USA / CURRENT RADAR

United States: labs, platforms, and infrastructure

A good part of the commercial and scientific frontier moves here. This block brings together all published radar news linked to U.S. or transatlantic sources, without limiting how many appear per lab.

30 OCT 2026 · NVIDIA DeveloperHow to create a Kubernetes runtime validated for NVIDIA Dynamo 30 OCT 2026 · AWS Machine LearningEnd-to-end management of Amazon Bedrock costs: playbook for FinOps teams and platform 30 OCT 2026 · AWS Machine LearningPreview live web apps while your code agent builds them with the Amazon Bedrock AgentCore runtime 30 OCT 2026 · NVIDIA DeveloperRun AI models to build apps at the edge with few commands 30 OCT 2026 · AWS Machine LearningWorldwide Express automates LTL freight exceptions with a multi-agent solution on Amazon Bedrock and reduces active cases by 75% 30 OCT 2026 · NVIDIA DeveloperNVIDIA NemoClaw: build specialized agents on a stack you control 30 OCT 2026 · Microsoft ResearchFrom synthetic data to real trust: evaluating Microsoft Copilot at enterprise scale, per Microsoft Research 30 OCT 2026 · OpenAIOpenAI will make it easier for websites to express AI training preferences via a shared signal in robots.txt 30 OCT 2026 · OpenAIOpenAI community safety report and new protections for teens 30 OCT 2026 · OpenAIOpenAI shares excerpts from court filings on how fair use supports AI training and innovation 29 OCT 2026 · AWS Machine LearningHow the UK Amazon Quick User Group grew in 2026: three quarterly meetups and over 150 attendees 23 OCT 2026 · AWS Machine LearningPractical guide: Multi-Turn Reinforcement Learning in Amazon SageMaker AI with Wordle 23 OCT 2026 · AWS Machine LearningBayer Crop Science develops ADE, an ‘agentic’ decision engine on Amazon Bedrock AgentCore with AWS Unified Operations 23 OCT 2026 · NVIDIA DeveloperAI agents train robots like Microduck to walk in simulation and overcome obstacles 23 OCT 2026 · OpenAIOpenAI strengthens containment and safety in frontier model training: offline environments and high‑compute red teaming 23 OCT 2026 · OpenAIOpenAI and Poland’s NASK partner to advance AI safety and security 22 OCT 2026 · AWS Machine LearningAmazon Polly launches a major update to its Generative engine: more expressive, accurate voice, low latency, and 14 new voices in 9 locales; Hindi and Finnish in general availability, no code changes 22 OCT 2026 · AWS Machine LearningSageMaker AI adds training sessions to serverless model personalization, without managing GPUs; now generally available 22 OCT 2026 · AWS Machine LearningAmazon SageMaker introduces training session APIs to build Python LLM loops without managing GPUs 22 OCT 2026 · AWS Machine LearningAmazon Quick: how to deploy AI in HIPAA-eligible environments and manage PHI flows 16 OCT 2026 · AWS Machine LearningTrust in AI agents: production-ready evaluation framework with Amazon Bedrock AgentCore 16 OCT 2026 · OpenAIItaú applies an AI First strategy with ia.i and IARA for its customer and employee base 16 OCT 2026 · NVIDIA DeveloperNVIDIA Nemotron 3.5 Super VL powers specialized agents with multimodal reasoning 16 OCT 2026 · NVIDIA DeveloperNVIDIA RTX Spark: code, test, and deploy on a single machine 16 OCT 2026 · AWS Machine LearningReflection AI launches Beam: open‑weights model for agentic programming on AWS 16 OCT 2026 · OpenAIStack Overflow for Agents triples sign-ups; its ChatGPT plugin boosts agent creation up to 10× 15 OCT 2026 · NVIDIA DeveloperNVIDIA Run:ai is now open source as part of the NVIDIA DSX AI Factory platform 15 OCT 2026 · NVIDIA DeveloperMeta launches Muse Spark Open, an MoE model to scale agentic coding workflows on NVIDIA GB300 NVL72 15 OCT 2026 · NVIDIA DeveloperReplaySSM rethinks state storage to speed up decoding in hybrid models 09 OCT 2026 · AWS Machine LearningSeptember 2026 updates for AI creators: Amazon Bedrock, Amazon Bedrock AgentCore, and Strands expand models, speed up serverless agents, and automate knowledge base syncing 09 OCT 2026 · AWS Machine LearningHow Postman runs Agent Mode for 40 million developers on Amazon Bedrock 09 OCT 2026 · OpenAIAsana cuts model costs 76x in browser tests with GPT-6.1 Sol 08 OCT 2026 · NVIDIA Developer5 steps to create SimReady assets for robotics with Frontier AI Models 08 OCT 2026 · AWS Machine LearningPayment for inference in AI agents: BlockRun and Incarna use Amazon Bedrock AgentCore payments 08 OCT 2026 · NVIDIA DeveloperNVIDIA KGMON, second in KDD Cup 2026 Data Agents: lessons for reliable analytics agents 08 OCT 2026 · AWS Machine LearningReference architecture: share GPU clusters on EKS with Amazon SageMaker HyperPod, with team isolation and fairness 08 OCT 2026 · OpenAILegalOn cuts Codex’s estimated daily costs by 65% and keeps development on pace 08 OCT 2026 · Apple Machine LearningApple Machine Learning introduces Normalizing Trajectory Models for few-step generation with exact likelihood 07 OCT 2026 · AWS Machine LearningClaude Haiku 5.5 arrives on Amazon Bedrock and Claude Platform on AWS: the fastest and most efficient in the family, according to Anthropic, and costs about 75% less than Claude Haiku 4.5 for most tasks 02 OCT 2026 · OpenAIGPT-6 family model guide for startups: selection, reasoning tuning, and production readiness 02 OCT 2026 · AWS Machine LearningAnalyze thousands of leases for compliance with Amazon Quick and the Adjudicated Query pattern 02 OCT 2026 · AWS Machine LearningAdd safe web search to Claude Desktop with Amazon Bedrock AgentCore 02 OCT 2026 · AWS Machine LearningFine-tuning a search agent with multi-turn reinforcement learning on Amazon SageMaker AI 02 OCT 2026 · OpenAIChatham Financial scales its capital markets expertise with OpenAI 02 OCT 2026 · Apple Machine LearningStrengthening language discrimination improves language learning in multilingual voice models 01 OCT 2026 · AWS Machine LearningAWS Professional Services scales cloud migrations with agent-based AI on Amazon Bedrock AgentCore 01 OCT 2026 · AWS Machine LearningAmazon Quick: Live Data in Apps lets you serve governed real-time data in AI-built apps with Quick Sight 28 SEP 2026 · NVIDIA DeveloperNVIDIA Open Agent Safety Platform, a reference for continuous in-silicon agent monitoring 28 SEP 2026 · NVIDIA DeveloperNVIDIA OpenShell adds runtime controls to AI agents 27 SEP 2026 · NVIDIA DeveloperNVIDIA DSX MaxLPS, to maximize performance and efficiency in AI factories 26 SEP 2026 · NVIDIA DeveloperNVIDIA Developer introduces NV-Reason-CT, an open 3D TC VLM for chain-of-thought reasoning in radiology 25 SEP 2026 · OpenAIProaction boosts sales 60% and saves over 75 hours with Codex 25 SEP 2026 · AWS Machine LearningSpeed up multimodal RL training with SkyRL on Amazon SageMaker HyperPod 25 SEP 2026 · AWS Machine LearningNarrateAI brings LLM quality assurance to production on Amazon Bedrock with streaming evaluation and around 99% numerical accuracy 25 SEP 2026 · AWS Machine LearningQwen3‑TTS on Amazon SageMaker AI: real‑time custom voice and cross‑language cloning 25 SEP 2026 · AWS Machine LearningDatacor enables self‑service rental analytics for gas and welding distributors with Amazon Quick Sight on TrackAbout 24 SEP 2026 · AWS Machine LearningTranscription with speaker labels using WhisperX on Amazon SageMaker AI 24 SEP 2026 · AWS Machine LearningMulti‑account architecture for AI agents with Amazon Bedrock AgentCore Gateway and MCP 18 SEP 2026 · AWS Machine LearningAmazon SageMaker Inference: recap of 13 launches in 2026 to date 18 SEP 2026 · NVIDIA DeveloperLLM inference benchmarking at scale with AIPerf 18 SEP 2026 · AWS Machine LearningMoonshot AI’s Kimi K3 arrives on Amazon Bedrock: open-weights option for code and knowledge, with native vision and 1M-token context 18 SEP 2026 · AWS Machine LearningMigration of multimodel AI agents to the Amazon Bedrock AgentCore runtime from Amazon ECS with AWS Fargate, maintaining three-model orchestration and knowledge retrieval with vectors 18 SEP 2026 · AWS Machine LearningAmazon introduces the new AgentCore runtime: elastic, optimized, and with consistent cold starts 18 SEP 2026 · AWS Machine LearningDeploy Hugging Face models on Amazon SageMaker AI with six open-source agent skills 18 SEP 2026 · AWS Machine LearningAmazon SageMaker HyperPod Inference Gateway: native Kubernetes add-on for Amazon EKS with GPU-aware routing 18 SEP 2026 · OpenAIOpenAI launches the Australian Youth Safety Blueprint: a six‑pillar roadmap for safer AI for young people 18 SEP 2026 · Apple Machine LearningDynamically Scaled Activation Steering (DSAS): dynamic steering adjustment in generative models 17 SEP 2026 · NVIDIA DeveloperNVIDIA speeds up dropless MoE training in JAX with Transformer Engine 17 SEP 2026 · AWS Machine LearningAmazon Connect Talent, AI solution to speed up hiring with consistent interviews and assessments 14 SEP 2026 · OpenAIPerplexity relies on GPT-6 Astra to run end-to-end systems 11 SEP 2026 · AWS Machine LearningProduction agent lifecycle oversight with AWS DevOps Agent and Amazon Bedrock AgentCore Evaluations 11 SEP 2026 · AWS Machine LearningBeyond price per token: choosing the OpenAI model on Amazon Bedrock with outcome metrics 11 SEP 2026 · AWS Machine LearningBuild and deploy interactive MCP apps with Amazon Bedrock AgentCore 11 SEP 2026 · Apple Machine LearningSimpleDesign: joint model for co-designing protein sequences and structures 11 SEP 2026 · Apple Machine LearningVideo descriptions A/B tested: quality measured with multiple-choice questions 11 SEP 2026 · Apple Machine LearningDiscoSign: text-to-gloss translation in sign language with discourse awareness via a modular LLM framework 10 SEP 2026 · AWS Machine LearningAmazon SageMaker HyperPod adds model cache to speed up cold starts for inference 10 SEP 2026 · AWS Machine LearningVideo and image search in Amazon Bedrock Knowledge Bases with Marengo Embed 3.0 by TwelveLabs 10 SEP 2026 · AWS Machine LearningAmazon Quick reaches general availability on desktop for macOS and Windows; adds activity feed on iOS and Android 10 SEP 2026 · NVIDIA DeveloperHow NIM Full‑Stack optimizations achieve 2.5x more users on Nemotron 3 Ultra 06 SEP 2026 · OpenAIInside OpenAI, code agents are accelerating AI research 04 SEP 2026 · NVIDIA DeveloperNVIDIA introduces CUDA Rust: two paths to write GPU kernels in Rust 04 SEP 2026 · AWS Machine LearningHow to deploy a multimodal ordering assistant on WhatsApp with Amazon Bedrock AgentCore and Amazon Nova 2 04 SEP 2026 · NVIDIA DeveloperAlibaba releases Qwen3.8-Flash-Next weights for Agentic Coding tests on NVIDIA GB300 NVL72, as a preview of Qwen4 04 SEP 2026 · NVIDIA DeveloperHow to build a memory-powered AI agent with NVIDIA NemoClaw for enterprise environments 04 SEP 2026 · AWS Machine LearningLifecycle policy design for AgentCore memory in Amazon Bedrock: scoring, consolidation, and pruning with AWS Step Functions and AWS CDK 04 SEP 2026 · NVIDIA DeveloperAdvanced edge reasoning: how to deploy and optimize models on NVIDIA Jetson 04 SEP 2026 · AWS Machine LearningBuild a Physical AI model factory with NVIDIA Cosmos 3 on SageMaker HyperPod on Amazon EKS, with GPU goodput as a key metric 04 SEP 2026 · AWS Machine LearningHyperPod InstantStart integrates Amazon EKS and SageMaker HyperPod for agent-driven operations 04 SEP 2026 · AWS Machine LearningCustomize your knowledge base in Amazon Bedrock for long, complex documents with Amazon Textract 04 SEP 2026 · AWS Machine LearningIntuit creates EWOK Agent, an agentic assistant for disaster recovery on Amazon Bedrock 01 SEP 2026 · NVIDIA DeveloperHow to run local agent flows with Meta’s Muse Glimmer and NVIDIA 20 AUG 2026 · NVIDIA DeveloperNVIDIA Nemotron 3.5 Lightning accelerates specialized tasks for long-running agents 20 AUG 2026 · NVIDIA DeveloperHow to distribute agent work across different models with NVIDIA NeMo Switchyard 10 AUG 2026 · AWS Machine LearningInteractive development environments on Amazon EKS with SageMaker AI 10 AUG 2026 · AWS Machine LearningHow nOps cut their FinOps agent development time with Amazon Bedrock AgentCore 10 AUG 2026 · OpenAIModel ML uses GPT-5.6 Sol to streamline financial work 10 AUG 2026 · OpenAIAdvanced cybersecurity models in the hands of authorized partners 10 AUG 2026 · OpenAIOpenAI expands Daybreak for cybersecurity research and testing 07 AUG 2026 · AWS Machine LearningHow Cohere Health digitizes clinical policies with Amazon Bedrock AgentCore 07 AUG 2026 · Apple Machine LearningScaling categorical flow maps for language models

METHOD NOTE The United States is fed automatically from news published and reviewed in Spanish. When an organization doesn’t offer a stable feed—common with public Chinese sources—its context is reviewed editorially and the primary link is preserved.

/ CROSS-CUTTING GOVERNANCE

Reliable data and automations under control.

To govern means defining owners, boundaries, and evidence before relying on a system.

01

Data governance

We agree on which information is reliable, who can access it, how long it’s kept, and how it’s corrected.

02

Agent governance

We define what an automation or agent can do, when it requires approval, and how its actions are reviewed.

See how we apply it
WhatsApp