Google Cloud introduced the Gemini agent on Oct. 8 at Gemini at Work 2026, describing the new workplace AI system as a “single, universal agent for work.”
According to Google Cloud CEO Thomas Kurian’s keynote write-up, Gemini can answer questions, handle knowledge work, create images and media, and write and run code from a single prompt box. The agent is designed to plan work, use company-approved skills and tools, connect to business systems, and return finished work inside documents, inboxes and developer environments.
Google said Gemini can be accessed from web, iOS, Android, Windows and Mac, as well as channels including the command line, Google Workspace, Microsoft 365 and Slack. It can also be integrated into third-party applications or run without a dedicated user interface as a headless agent.
The release builds on Gemini Enterprise Agent Platform, which Google Cloud launched in April 2026 as its enterprise layer for building, scaling, governing and optimizing agents.
Why the coworker model stands out
In Workspace, a team can describe the role it needs and Gemini creates a coworker agent. Google said that agent receives a Workspace account, email address, calendar, Drive storage and company directory presence. The agent can then be added to a Chat space, mentioned in a document comment or asked to post completed work back to a group.
That design makes the announcement more significant than another AI sidebar. A chatbot usually borrows a person’s context and acts inside one session. A persistent coworker agent has its own identity, its own audit trail and access limited to what a team shares with it. For IT teams, that turns AI adoption into an identity-and-permissions problem as much as a productivity problem.
How Gemini is supposed to work across a company
Google described the new agent around several capabilities: unified chat and autonomous task execution, persistent cloud-based memory, multi-agent orchestration, reusable skills, system connections and flexible model choice.
Persistent execution: Gemini runs in the cloud, so tasks can keep running after an employee closes a laptop. Google said work that takes hours or days can continue and remain available when the employee returns.
Multi-agent orchestration: Gemini can create temporary, job-specific sub-agents with their own identities to work through complex tasks in parallel or sequence.
Skills and tools: Gemini can connect to workplace tools such as Confluence, Microsoft Office, Teams, Slack, Workspace, Git, Jira, Salesforce, ServiceNow, BigQuery, Databricks, Postgres and Snowflake, according to Google. It can also connect through Model Context Protocol servers.
Model choice: Google said Gemini can route work across its Gemini models and Anthropic’s Claude models today, with other private and open models planned for the future.
Governance is central to Google’s pitch
Google’s announcement spends considerable time on governance because autonomous agents can do more damage than a text generator if they are over-permissioned. The company said every agent gets its own cryptographically attested identity, least-privilege permissions and audit logs that attribute actions to the agent rather than to a person.
Access controls can map through standards such as OAuth when Gemini connects to external systems. Google also described Agent Sandbox and Agent Gateway: Agent Sandbox isolates task execution, while Agent Gateway works as an AI network firewall that enforces company policies across agent traffic.
Workspace becomes a control surface
For Google Workspace customers, the agent is designed to work inline in Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar. Google gave examples such as coordinating a meeting from prior event context, creating a financial model in Sheets after research, then building a Slides deck from the results.
The same agent can proactively suggest tasks based on incoming messages. In Google’s example, Workspace Intelligence can recognize a manager’s email asking for a project update deck and offer to delegate that work to Gemini.
Google is also pushing industry-specific agents
Gemini for Financial Services and Gemini for Legal are in preview, according to Google Cloud. The financial services version draws on sources such as FactSet, LSEG, S&P Global, SEC filings and proprietary repositories, and Google said it includes more than 50 foundational skills. The legal version is designed to inherit matter-level permissions and ethical walls from document management platforms including NetDocuments and iManage.
Google said Government, Healthcare and Retail versions are coming soon. Regulated industries are likely to judge workplace agents less by demo quality and more by permissions, data lineage, auditability and whether sensitive data can stay inside approved systems.
Customer examples show where Google wants this to go
Google listed several organizations using Gemini Enterprise, including BNP Paribas, Bradesco, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty and Wesfarmers. The company said Orange Spain has deployed more than 1,000 custom Gemini Enterprise agents across HR, IT, sales and customer service. It also said Bradesco used Gemini Enterprise for fiscal, accounting and contractual analysis, cutting document review from one hour to five minutes and reducing risk inconsistencies by 60%.
Those examples are Google-reported customer results, not independent benchmarks. They still show the business category Google is targeting: AI systems that run repeatable work across departments, not only individual productivity features.
What remains unclear
The announcement does not settle the practical questions businesses will face during rollout. Google outlined capabilities and controls, but real deployments will still depend on how each organization configures access, tests agent behavior, limits spending and decides which tasks can run without human approval.
The Gemini agent is currently in private preview for enterprise customers. Google says wider availability is coming soon for Workspace customers on select Business and Enterprise plans, but it has not announced a general-release date or public pricing.
How smaller teams should govern workplace agents
Small and midsize teams do not need to copy an enterprise rollout. Workplace agents should be treated as governed software identities. The moment an AI tool can access documents, write code, coordinate meetings or trigger workflows, it needs the same discipline applied to employees, contractors and service accounts: scoped permissions, logs, approval paths, budgets and clear ownership.
Get new small business insights by email
Practical ideas and useful articles to help you make better business decisions.
Tech Help Canada Staff researches, writes, and reviews practical content for business owners and professionals. Our coverage spans business, marketing, SEO, technology, and the tools and systems people use to grow and operate online. We focus on clear, useful information backed by research, hands-on experience, and editorial review. Learn more about our team and editorial standards. Need help with something? Contact Us
