Overview
- Google unveiled the Gemini agent on Oct. 8, 2026, as a cloud‑native, persistent AI that accepts objectives, runs multi‑step workflows and returns finished work to apps like Gmail, Docs and Sheets.
- Companies can create 'coworker agents' that have their own Workspace email, calendar, Drive storage and directory listing so agents act as distinct, auditable participants in workflows.
- Gemini maintains four memory types (session, semantic, procedural, episodic) and spawns temporary sub‑agents to run tasks that can last hours or days without losing context.
- The platform routes jobs to the best model for each task, currently using Google’s Gemini models and Anthropic’s Claude, and offers spend caps and smart routing to control token costs.
- Google is rolling the agent out in private preview to select Workspace Business and Enterprise customers with financial‑services and legal versions available now and other industry packs coming soon while pricing and general availability remain undisclosed.