# News: October 21, 2026

> 11 curated links on AI agents, Python, security, Google Cloud, and engineering, selected by Médéric Hurier.

- Author: [Médéric Hurier (Fmind)](https://www.fmind.dev/)
- Published: 2026-10-21
- URL: https://www.fmind.dev/news/2026-10-21/

- [Claude Opus 5.5 and Claude Sonnet 5.5 available in Antigravity](<https://cloud.google.com/blog/products/ai-machine-learning/claude-opus-55-and-claude-sonnet-55-available-in-antigravity>): Google Cloud brings Claude Opus 5.5 and Claude Sonnet 5.5 to Google Antigravity across its CLI, IDE extensions, and Antigravity 2.0 runtime for enterprise agentic workflows.
- [Voxtral 3](<https://mistral.ai/news/voxtral-3>): Mistral AI releases Voxtral 3, featuring advanced speech-to-text and text-to-speech models alongside a real-time Voice Agent SDK supporting live tool calling and low-latency audio pipelines across 50+ languages.
- [Harness²](<https://github.com/google-research/harness2>): Google Research open-sources Harness², a plug-and-play framework for recursive agent harnessing that dynamically refines prompts, tools, memory, and orchestration while keeping model weights fixed.
- [Bringing uv and ty together](<https://astral.sh/blog/uv-ty-integration>): Astral introduces experimental integration between uv and ty, enabling first-class support for Python scripts on the CLI and in editors alongside a unified uv check command.
- [Introducing Agent Memory on Databricks](<https://www.databricks.com/blog/introducing-agent-memory-databricks>): Databricks introduces Agent Memory to enhance AI agent accuracy while decreasing latency and costs by persisting critical facts, preferences, and multi-step workflows.
- [Devs guide to MCP Apps in Gemini](<https://cloud.google.com/blog/topics/developers-practitioners/devs-guide-to-mcp-apps-in-gemini>): Google Cloud publishes an implementation guide detailing how developers can build and integrate Model Context Protocol (MCP) applications into Gemini to support visual and interactive multi-step workflows.
- [Build a voice agent that never goes silent using ADK](<https://developers.googleblog.com/step-by-step-guide-build-a-voice-agent-that-never-goes-silent-using-adk>): Google publishes a step-by-step tutorial on building responsive voice agents that never go silent using open-source ADK, streaming tools, and Gemini 3.8 Live with response scheduling.
- [NVIDIA GB300 FP4 vs TPUv7 FP8 performance per dollar](<https://x.com/SemiAnalysis_/status/2112258243372146742>): SemiAnalysis reports that NVIDIA's first GB300 FP4 submission remains 1.64x worse in performance per dollar compared to Google's TPUv7 FP8 submission.
- [Google named a Leader in 2026 IDC MarketScape for Agentic AI Platforms](<https://cloud.google.com/blog/products/ai-machine-learning/google-named-leader-in-2026-idc-marketscape-for-agentic-ai-platforms>): Google Cloud is recognized as a Leader in the inaugural IDC MarketScape: Worldwide Agentic AI Platforms 2026 Vendor Assessment for its connected Gemini foundation across developer and enterprise surfaces.
- [Key Takeaways from Gemini at Work 2026](<https://www.youtube.com/live/V_z-xV5b1B8>): Google Cloud summarizes enterprise highlights from Gemini at Work 2026, focusing on portfolio brand unification under Gemini, cross-stack enterprise agent rollouts, and predictable compute strategies.
- [Use Sheets canvas to plot raw feature requests onto an interactive 2x2 effort vs. impact matrix](<https://www.youtube.com/watch?v=o0gMKOIGdQo>): Google Workspace demonstrates how teams can use Sheets canvas to plot raw feature requests directly onto an interactive 2x2 effort-versus-impact prioritization matrix.