At its Gemini at Work 2026 event on October 8, 2026, Google Cloud CEO Thomas Kurian introduced the Gemini agent — which Google describes as "a single, universal agent for work" that answers questions, handles knowledge work, creates media, and writes and runs code. The agent runs on the web, iOS, Android, Windows, Mac, and the command line; inside Google Workspace, Microsoft 365, and Slack; and headlessly inside third-party applications. The pitch: one interface, one API, and an objective instead of step-by-step instructions.
The design centers on persistent execution. The agent runs in Google's cloud with a shared memory graph — session, semantic, procedural, and episodic — so a job that takes hours or days continues after the user closes a laptop. It spins up temporary sub-agents for individual jobs, each with scoped identity and permissions, and it supports persistent coworker agents that hold their own @agents.company.com email address, calendar, Drive storage, and company-directory seat. Colleagues can add one to a Chat space, mention it in a document, or assign it objectives like any teammate — while it sees only what is explicitly shared with it.
The agent draws on a tools registry including Salesforce, ServiceNow, Jira, Snowflake, Databricks, Microsoft Office and Teams, and any Model Context Protocol server, plus a skills registry of reusable workflows. Critically, it is multi-model: Google routes each job across its own Gemini family and Anthropic's Claude today, with other private and open models planned. A Smart Routing layer picks the model per job, and administrators can set real-time, per-project spend caps that pause an agent when triggered.
Governance is where Google is differentiating. The model rests on cryptographically attested agent identity, OAuth-propagated permissions, action-level audit trails attributed to the agent rather than a human user, an Agent Sandbox, and an Agent Gateway Google calls an AI network firewall. On the data side, Google introduced a Knowledge Catalog for business definitions, Smart Storage for unstructured-data enrichment, and a Borderless Lakehouse that can query Amazon S3 and Azure Data Lake without variable egress fees. Industry-specific agents for financial services and legal are in preview, with government, healthcare, and retail to follow.
Two things Google did not announce: pricing and a general-availability date. The agent is in private preview for enterprises, and pricing details have not been published. For enterprise IT, the test plan writes itself: validate the identity model against your own IAM, test the spend caps against a runaway loop, measure what the multi-model routing actually saves, and confirm the audit trails satisfy your compliance regime. The vendors' assistant wars have moved from chatbots to coworkers with directory seats — and the security architecture will decide who wins.