# News: October 22, 2026

> 7 curated links on AI agents, Python, security, Google Cloud, and engineering, selected by Médéric Hurier.

- Author: [Médéric Hurier (Fmind)](https://www.fmind.dev/)
- Published: 2026-10-22
- URL: https://www.fmind.dev/news/2026-10-22/

- [How NVIDIA OpenShell reference architecture secures and governs AI agent workloads at scale](<https://developer.nvidia.com/blog/how-nvidia-openshell-reference-architecture-secures-and-governs-ai-agent-workloads-at-scale>): NVIDIA open-sources OpenShell, a reference architecture combining Gateway, Supervisor, Sandbox, and Policy Prover components to enforce hardware-isolated security, governance, and rapid quarantine for agentic AI workloads at scale.
- [Introducing Claude Duet](<https://www.youtube.com/watch?v=Fb-68zqcSVs>): Anthropic introduces Claude Duet, an autonomous agent application that surfaces priorities, executes workflows in the background after the user puts their phone down, and verifies sensitive actions before proceeding.
- [Powering the Agent Development Lifecycle with any framework, from ADK to LangChain and beyond](<https://developers.googleblog.com/powering-the-agent-development-lifecycle-with-any-framework-from-adk-to-langchain-and-beyond>): Google introduces the Agents CLI Extension System, enabling developers to scaffold, evaluate, deploy, and enforce compliance policies across multiple agent frameworks including LangChain within a unified command surface.
- [Google Cloud infrastructure at NVIDIA GTC Berlin 2026](<https://cloud.google.com/blog/topics/ai-infrastructure/google-cloud-infrastructure-at-nvidia-gtc-berlin-2026>): Google Cloud and NVIDIA expand their infrastructure collaboration with A5X instances powered by the Vera Rubin NVL72 platform, Virgo Network scaling, Managed Lustre GPUDirect Storage, and sovereign AI capabilities.
- [Same GPU, 50% More Output — Why the Operator Matters](<https://www.youtube.com/watch?v=3axY1wz2OvA>): SemiAnalysis explains how optimized serving stacks deliver 50% more output on identical B200 hardware while introducing the ClusterMAX benchmark to evaluate cluster operator efficiency.
- [Sundar Pichai on recursive self-improvement, compute dilemmas, and future interfaces](<https://www.youtube.com/watch?v=F3ItSTZNtT0>): Alphabet CEO Sundar Pichai discusses Google's frontier AI roadmap, compute allocation tradeoffs between research and cloud revenue, recursive self-improvement, and next-generation agent interfaces on the Big Technology Podcast.
- [How an AI hedge fund lost $35 billion in 72 hours](<https://www.youtube.com/watch?v=Tms3kCQiwQo>): The Infographics Show analyzes how Leopold Aschenbrenner's AI-focused hedge fund faced margin calls and forced liquidations following massive leveraged bets on AI infrastructure and semiconductor stocks.