Overview
- Google announced the Gemini agent on Oct. 8, 2026 as a private‑preview product for enterprise customers with industry previews now available for finance and legal and broader rollouts planned.
- Gemini agents can act as persistent coworkers with their own Workspace identities, company email addresses, inboxes and storage so they can accept objectives and carry work across days.
- The system maintains four types of memory—session, semantic, procedural and episodic—and can run multi‑step workflows and sub‑agents for up to seven days to produce finished documents, code or meetings.
- Under the hood the platform chooses the best model for each job, currently using Google’s Gemini models and Anthropic’s Claude, and offers developer tools, a Gemini Enterprise app, APIs and cost controls like smart routing and real‑time spend caps.
- Google built cryptographic agent identities, audit trails and scoped access to address enterprise security and compliance needs but the staged rollout will test how regulators, IT teams and workers adapt to always‑on autonomous agents.