Google folded its work agents into one: Gemini agent

Google Cloud announced the Gemini agent on October 8 at Gemini at Work 2026. From a single prompt box it answers questions, handles knowledge work, creates images and media, and writes and runs code. The agent keeps the same memory and controls everywhere it runs: Gmail, Drive, Docs, Slides, Sheets, Chat, Calendar, plus third-party services. Google’s pitch is that you hand it objectives rather than instructions. Google Cloud Blog Google 9to5Google

Google says the agent runs each job on the model that fits best, drawing on both the Gemini and Claude families today. The launch also carries cost controls, including Smart Routing and real-time spend caps. Access starts with enterprise customers, the Gemini Enterprise app and Workspace admin settings. TechCrunch

Your real competitor is the platform’s agent

The news is not that Google has an agent. It is the shape. Once an assistant holds your organization’s context and acts inside the tools you already use, a chat box that reads your data stops being a product and becomes a feature of the surface you sit on. Google routing between Gemini and Claude per job points the same way. Baking one model into your code is the risky bet now. What you keep is your own context and workflow, plus the connectors that put your data where the agent already lives.

For a one-person team that leaves two things to check. If your product is a thin layer over a single model, the platform’s default agent takes that spot first. And because access opens from enterprise customers and admin settings downward, the room left for an independent developer sits in connectors and specialized workflow. Permissions and agent identity are enterprise features for now, and how far they reach into personal accounts is still open.

The rest of today’s news

  • Australia signals frontier AI safety rules: the government said on October 8 that it is drafting legislation requiring AI companies to prove they can identify and manage risks, taking cues from banking and aviation regulation. ABC News
  • USA Today sues OpenAI for $250 million: USA Today’s parent filed in Manhattan federal court on October 8, alleging OpenAI trained its models on hundreds of thousands of articles without permission. OpenAI has not yet responded. The Verge
  • Harness acquires Augment Code assets: on October 8 Harness picked up Cosmos, the Auggie CLI, the Code Context Engine and the team behind them. Harness
  • JetBrains opens Mellum2.1 for coding agents: an Apache 2.0 12B mixture-of-experts model with 2.5B active parameters, built to run a coding subagent on your own hardware. JetBrains
  • China’s Manus raises over $500M: its first round since the collapsed Meta deal, led by Boyu Capital and IDG Capital. TechCrunch
  • OpenAI brings GPT-6 to the free tier: Intelligent UI turns answers into charts, buttons and forms, and free and Go users get GPT-6 Luna from October 8. TechTimes
  • Odyssey unveils the Odyssey-3 world model: the lab says one model drives robot arms, humanoids, cars and drones, and opened a research preview. TNW