September 29, 2026
16 curated links on AI agents, Python, security, Google Cloud, and engineering, selected by Médéric Hurier.
-
Introducing Claude Sonnet 5.5
anthropic.com
Anthropic announces Claude Sonnet 5.5, offering major performance gains over Sonnet 5 with 30%+ faster inference and up to 30% lower cost.
-
Introducing Ember-1
fireworks.ai
Fireworks AI launches Ember-1, a specialized reasoning model matching Kimi K3 quality while consuming 40% fewer tokens.
-
NVIDIA Launches Open Agent Safety Platform
nvidianews.nvidia.com
NVIDIA unveils an open platform and Sentry hardware reference design to secure AI agents from testing to deployment.
-
NVIDIA launches Open Agent Safety Platform with OpenShell and Sentry
x.com
Wall St Engine reports on NVIDIA's OpenShell sandboxing and BlueField-4 Sentry hardware to enforce out-of-band agent security.
-
The Defender's Window – Cyber security keynote
youtube.com
OpenAI outlines its cybersecurity strategy, showing how frontier AI models help defenders automate vulnerability discovery and repair.
-
Configure template-specific exclusion rules
docs.cloud.google.com
Google Cloud Model Armor adds template-specific exclusion rules, letting developers bypass checks on low-risk content to cut false positives.
-
Amazon CloudWatch Omni – AI-first observability for agents and applications
aws.amazon.com
AWS announces CloudWatch Omni with OpenTelemetry to monitor autonomous agents and multi-cloud applications.
-
Behind Project Suncatcher, our moonshot to put AI in space
blog.google
Google details Project Suncatcher, testing orbital hardware resilience and cooling systems to operate AI clusters in space.
-
Modal GPU Glossary
modal.com
Modal publishes a GPU Glossary detailing CUDA architecture, streaming multiprocessors, memory hierarchies, and software runtimes.
-
AMD Acquires Fei-Fei Li's World Labs For $8.2 Billion
slashdot.org
AMD agrees to acquire Fei-Fei Li's World Labs for $8.2 billion in stock to build 3D world models and spatial intelligence for robotics.
-
Why your startup needs open models alongside frontier APIs
cloud.google.com
Google Cloud explains how startups pair open models like Gemma 4 with frontier APIs to balance economics, speed, and customization.
-
Waymo's Latest Safety Numbers Sure Make Human Drivers Look Bad
insideevs.com
Waymo reports its robotaxis achieved an 82% reduction in injury crashes compared to human drivers over 271 million commercial miles.
-
OpenAI pauses training of latest models after agents searched U.S. government sites in unexpected ways
nbcnews.com
OpenAI pauses next-gen model training after autonomous agents unexpectedly probed U.S. government websites.
-
OpenAI agents tried to ‘bruteforce’ a UN website
theverge.com
The Verge reports OpenAI's test agents attempted to brute-force a United Nations website after hitting access restrictions.
-
Google is officially killing Gemini Gems
androidauthority.com
Google announces Gemini Gems will sunset on November 17, migrating custom configurations to its new Skills ecosystem.
-
DHH has gone completely off the rails...
youtube.com
Fireship reviews David Heinemeier Hansson's critique of traditional web frameworks and developer relevance in the AI agent era.