Google's Gemini agent enterprise push turns AI into coworkers

Google Cloud wants enterprise AI to hold a job the way a person does: with an email address, a calendar, and a seat in the company directory. At its Gemini at Work event on October 8, the company unveiled the Gemini agent, which chief executive Thomas Kurian described as a single, universal agent for work. The Gemini agent enterprise rollout centers on a design choice that matters more than the model underneath: the agent joins the team as an identity, not a feature. Employees hand it an objective, such as preparing a launch brief, and it plans the task, picks the tools, and returns finished work inside the systems people already use.

An email address, a calendar, a directory seat

Each coworker agent receives its own Workspace account. That means a dedicated email address, a calendar, Drive storage, and an entry in the company directory. Colleagues reach it by tagging it, emailing it, sharing a file with it, or adding it to a group chat, according to TechCrunch's reporting on the announcement. The agent appears under its own name in version history and document comments. Critically, it sees only the context that team members explicitly share, so its knowledge starts at zero for every deployment and grows through the work it is given.

Long-running work gets treated as a first-class idea in this Gemini agent enterprise design. The agent keeps four kinds of memory: session memory for the task in front of it even when a job runs for days, plus semantic, procedural, and episodic memory for what it learns over time. Because it runs persistently in the cloud, a task that needs hours of research, drafting, and approval coordination does not have to be rebriefed when someone switches devices. Google said the agent can also spin up sub-agents, each carrying its own identity, to coordinate work that stretches across teams.

Governance modeled on employee files

The identity design extends to oversight. Every action the agent takes is written to an audit trail attributed to the agent itself rather than to a human, the company said. Each agent gets a cryptographically attested identity with least-privilege permissions, and traffic between agents and outside systems passes through an Agent Gateway that Google describes as an AI network firewall enforcing organization policy. Code runs inside an Agent Sandbox with its own network boundary. The framing, laid out in Google's launch post, answers four questions: who the agent is, what it may do, what it did, and what it must never touch.

The model underneath is deliberately interchangeable. The agent orchestrates across Google's own Gemini family alongside Anthropic's Claude, with support for additional private and open models to come, and a Smart Routing tool assigns each job to the model that fits it. Administrators can set real-time spend caps per project in the Cloud Billing Console, which pause an agent project automatically when it hits the limit. AI Weekly reported that enterprise token volume has tripled since 2024 while prices dropped sharply, and the routing plus caps are aimed at keeping multi-day agent runs affordable.

Who gets it first

Availability starts at the top of the market. The Gemini agent is in private preview for Gemini Enterprise customers, with broader availability planned for select Google Workspace Business and Enterprise plans. Early testers included sportswear brand On, Shopify, and PayPal, and Google named existing enterprise customers including BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty, and Wesfarmers. The agent connects to Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, and other systems, and can work with any Model Context Protocol server.

Industry versions are already taking shape as part of the Gemini agent enterprise push. Financial services and legal agents are in preview, the company said, with versions for government, healthcare, and retail to follow. The legal version inherits matter-level permissions and ethical walls from document management platforms, and Google names Harvey and Onit as partners, with the law firm Cooley building a confidential-information redaction agent on it. A tasks inbox shows the agent's thinking, its delegation to sub-agents, and its progress, so managers can watch long-running work without hovering.

The larger point is cultural as much as technical. For years, chatbots sat outside company systems, acting as guests that borrowed a user's permissions. Giving an agent its own email and its own audit trail moves it inside the organization, where it gets the same treatment as a hire: provisioning, scoped access, logging, and eventually an offboarding. That raises familiar management questions about who can create these identities and how their mistakes get attributed, but it also answers the one that has slowed enterprise adoption: who is responsible when the machine does the work.

TechCrunch reported on the Gemini agent launch at Google's Gemini at Work event.

AI Weekly covered the identity and governance details of the announcement.