Google Cloud on October 8 unveiled the Gemini agent at its Gemini at Work 2026 event, pitching it as a single, universal agent for work. In a keynote-adapted post on the Google Cloud Blog, CEO Thomas Kurian said work now "starts in the prompt window": employees hand the agent an objective rather than step-by-step instructions, and return to finished work.
The agent folds previously separate capabilities into one agent and one API: answering questions, handling knowledge work, creating images and media, and writing and running code. It can act as a personal assistant or as a team member — tracking a project on behalf of a group, or filling a role such as an analyst inside a finance department.
The design rests on "unified agent" plus "omnipresent access": web, iOS and Android, Windows and Mac, command line, Google Workspace, Microsoft 365, and Slack, plus embedding in third-party apps or running headless with no interface at all.
"You give it objectives, not instructions. You delegate an outcome and come back to finished work," Kurian wrote. Given an objective, the agent plans the work, invokes skills and tools, connects to the customer's business systems, and delivers the finished product inside the documents, inboxes, and developer environments people already use.
It selects the best model for each job, ships with built-in cost controls, and carries the security, administration, and governance controls enterprise customers demand — including identity and policy management, authorization controls, secure sandboxing, and network gateways.
The sharpest break from ordinary chatbots is persistent execution. The agent runs in the cloud with a single set of memories, context, and one personalization graph, so nothing is lost switching devices. Employees can assign work on one machine, close the laptop, and come back days later: multi-day workflows keep running, parallel assignments stay coordinated, and the agent responds to scheduled tasks and events.
For complex jobs it can assemble temporary teams of specialized sub-agents, each with a defined role and identity. Google described three Workspace modes: personal assistance (already briefed on your calendar, team, and projects), proactive delegation (Workspace spots delegatable tasks for one-click handoff), and a coworker mode in which the agent gets its own Workspace account — email address, calendar, Drive storage, and a presence in the company directory — acting under its own identity and seeing only what is shared with it.
Google says the agent's business context comes from three layers. A tools layer connects to Confluence, Office, Teams, Slack, Git, Jira, Salesforce, ServiceNow, databases including BigQuery, Databricks, Postgres, and Snowflake, files on a user's desktop, and any Model Context Protocol server; enterprises can publish tools to a company-wide registry. Skills are reusable bundles of instructions and workflows, with a global library plus team and personal registries. Memory comes in four flavors — session, semantic, procedural, and episodic — and the agent onboards itself "the way a new hire would," learning the user, the tools, and the team before starting work.
The event also brought companion announcements: data and analytics skills that let non-technical staff get operational answers from plain-language questions, industry specialization for financial services and legal teams, and multi-model orchestration — Gemini currently routes across its own model family and Anthropic's Claude models — with Smart Routing and real-time spend caps.
Kurian disclosed adoption figures: nearly 500 Google Cloud customers each processed more than one trillion tokens over the preceding year; nearly 80% of all Google Cloud customers are using its AI products; and nearly 90% of the Fortune 100 use Gemini Enterprise.
Named early users include sportswear brand On, which tested dynamic model selection; Shopify, blending frontier models for millions of merchants; PayPal, routing 10 million multi-model requests per week; Brazilian bank Bradesco, which cut document review time from one hour to five minutes; and Orange Spain, which has deployed more than 1,000 custom Gemini Enterprise agents. Google added that small and midsize businesses increased their usage of its AI tools more than fivefold year over year.
Gemini agent enters a crowded field — Microsoft's Copilot, enterprise offerings from OpenAI and Anthropic, Salesforce, and others are all competing to define the "AI coworker." Google's differentiation is deployment-shaped rather than model-shaped: cross-system access, long-running memory, controllable cost, and auditable governance.
For enterprise IT, the pitch is a digital worker that never clocks off and needs no onboarding. The risks are equally real: permission boundaries, data leakage, prompt injection, and accountability when an agent errs. Google is leading with governance to answer the second set of questions first. What comes next is the part no keynote can settle — how many companies trust it with core workflows, and whether the built-in cost controls hold up against real invoices.