Google Unveils Gemini Agent for Work, With Gmail, Drive and Coding Tools in One Place

Google has introduced a new Gemini agent for enterprise work, an ambitious attempt to turn the AI assistant into something closer to a digital coworker. Rather than just answering a question, Google says the agent can coordinate several steps, work across connected apps, write and execute code, and come back with a finished result.

Announced on October 8 at Gemini at Work 2026, the agent is meant to operate across Google Workspace, developer tools and third-party services. There is a major limitation: this particular new Gemini agent is currently available to enterprise customers in private preview, not as a feature that every personal Gmail user can enable immediately.

What is new about this Gemini agent?

The centerpiece is a single place to assign a job rather than switching between separate tools. Google describes a system that can plan tasks, choose tools, call other AI agents and keep the relevant work context available across devices.

In theory, you could ask it to assemble a project update from files and conversations, build a spreadsheet, draft a presentation and coordinate a meeting. Google says it can work inside Gmail, Drive, Docs, Sheets, Slides, Chat and Calendar while maintaining consistent context and controls.

That does not guarantee every task will be completed perfectly or without supervision. Companies will need to test what the agent actually does on their documents and enforce approval for important changes.

Read Google's official Gemini at Work announcement.

Google wants the agent to work beyond Workspace

The announcement isn't limited to Gmail and Docs. Google lists connections for Microsoft 365, Slack, Jira, Git, Salesforce, ServiceNow, BigQuery and other business systems, plus support for Model Context Protocol servers.

A single work request could therefore involve the knowledge base, issue tracker, code repository and analytics platform of a company. The quality of the result will depend on which connections an organization has enabled, what permissions the agent has and the reliability of the underlying tools.

Google also describes persistent cloud execution. A long-running task can continue after the user closes a laptop. That is different from keeping a browser tab open for a normal chatbot conversation.

It can create smaller agents to handle sub-tasks

For more complicated jobs, Google says the Gemini agent can coordinate temporary specialist agents. One might examine a dataset while another works on a document and a third prepares code.

There is also a coworker-agent concept, where an organization assigns an agent a specific role, its own identity and a dedicated email address. Google describes these agents as having persistent storage and limited access to information shared with them.

The practical attraction is clear: a team would not need to hand over a personal employee's entire account just so an automated assistant could coordinate a recurring process.

What about privacy, security and mistakes?

Allowing an AI agent to act across an inbox, company documents and external services creates much bigger risks than asking a chatbot to summarize text.

Google says enterprise administrators can assign individual identities and role-based permissions, inspect audit logs, and set network policies through its Agent Gateway. Code execution is designed to take place in controlled sandbox environments.

These are company-described security controls, not proof that the system cannot make mistakes, reveal information or carry out an unintended action. Businesses should still apply least-privilege access, human review and tests against their own workflows before trusting an autonomous agent with sensitive work.

One agent does not mean one AI model

Google says the agent can route tasks across different models, including models from Anthropic. A simpler operation might use a cheaper, faster model, while a complex request can go to something more capable.

The company also announced tools for monitoring expenses and setting project-level spending caps. That matters for businesses because a multi-step agent may use many model calls, integrations and cloud resources for a single request.

There is no confirmed universal Philippine retail price for this private-preview work agent. It should not be confused with the consumer Gemini subscription tiers or assumed to be included in an ordinary Google account.

Why this matters in the Philippines

Many Philippine offices already work in Google Workspace, Microsoft 365, cloud spreadsheets and shared reporting systems. An agent that moves between those systems could eventually reduce repetitive work such as compiling weekly reports, checking project statuses and preparing follow-up documents.

But enterprise adoption is not simply a matter of buying an AI subscription. IT teams must verify data governance, connector permissions, employee oversight, organizational availability and total operating costs.

For developers, Google's support for code tools and third-party integrations is potentially more important than another chatbot interface. It hints at workflows where software projects and business operations share the same automation infrastructure.

Can you use it today?

The new work agent is currently an enterprise private-preview offering, according to The Verge's coverage of Google's announcement. Google has not announced general public availability for every account or a universal consumer release date.

That makes this an important direction for workplace AI, but not a ready-to-use feature for every reader today. The key question to watch is how well the agent performs real multi-app tasks when access becomes broader.

Read Google's official overview and The Verge's availability report.