AI News Roundup: Agents Get a Workspace, Claude Edits Your Docs
The shift from chat windows to things that actually do work. Google's Gemini Agent for Work (announced October 8) runs as a persistent agent with its o
My job is to notice when a number is doing the talking. The shift from chat windows to things that actually do work.
Google's Gemini Agent for Work (announced October 8) runs as a persistent agent with its own Workspace account, email and calendar, and can be at-mentioned like a colleague. It even routes between Gemini and Claude depending on the task.
Anthropic opened a public beta on October 6 letting Claude create and edit Google Docs, Sheets and Slides from chat, with ask-before-edit as the default.
Liquid AI released open-weight decision models d1-3B and d1-omni-600M on October 7 for local deployment, returning structured decisions in one pass.
OpenAI's Dots, announced late September, put always-on agents in ChatGPT that each run on GPT-6 Astra with their own cloud computer and connect to more than 4,000 apps.
The evidence side: The pattern across all of it is that the labs are racing on price and usefulness now, not just benchmark scores.
Good enough to draft with, not to bet on.
Updated October 28, 2026: the launches below have all reached general users since.
Frequently asked questions:
Q: What is Gemini Agent for Work?
A: A persistent agent announced October 8 with its own Workspace account, email and calendar - it can be at-mentioned like a colleague and routes between Gemini and Claude depending on the task.
Q: Can Claude edit Google Docs now?
A: Yes - a public beta opened October 6 letting Claude create and edit Docs, Sheets and Slides from chat, with ask-before-edit as the default.
Q: What did Liquid AI release?
A: Open-weight decision models d1-3B and d1-omni-600M on October 7 for local deployment, returning structured decisions in a single pass.