Google Gemini Agent Can Get a Work Email and Run for Days

What is Google’s new Gemini agent?

Google’s Gemini agent is an AI system designed to handle ongoing work across business tools rather than answer a single prompt and stop. Google announced the product on October 8, 2026, and described it as a “single, universal agent for work” that can answer questions, handle knowledge work, create images and media, and write and run code. Google Cloud’s launch announcement sets out the company’s description of the product and its intended scope.

Google’s Gemini agent matters because the product is built to move between work surfaces while retaining memory, skills, and controls. Google says Gemini works inline in Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar, which could reduce the need to repeatedly copy instructions or context between applications. The limitation is that a shared AI context can be useful only when the information supplied to the agent is appropriate for the task.

Google’s new product is separate from a simple chatbot interaction because it is intended to perform work over time and across systems. Teams already considering rules for AI shopping agents should apply the same basic principle in workplace settings: an agent needs clear boundaries before it receives authority to act.

How do Gemini coworker agents receive work identities?

Gemini coworker agents can receive their own identity, an @agents.company.com email address, storage, and access to context that a user or team provides. Google says those persistent agents are designed to function as continuing collaborators instead of temporary sessions that disappear after one conversation. The dedicated identity can make an agent easier to recognize in a work process, but it also means organizations must decide what the identity is allowed to access.

Google’s email-address feature does not mean an AI agent should automatically receive the same permissions as a human employee. An email identity can be useful for receiving task-related messages or operating within a defined workflow, while broad mailbox, file, or calendar access could expose information outside the task’s actual need. The most sensible approach is to start with the smallest practical set of data sources and permissions.

The work-email feature has also been described in independent launch coverage, although Google’s own announcement remains the primary source for the product details. Independent coverage of the Gemini agent launch highlights the unusual step of assigning an AI worker a dedicated email identity.

Where can the Google Gemini agent work?

Google’s Gemini agent is available through web, iOS, Android, Windows, Mac, command line tools, Google Workspace, Microsoft 365, and Slack, according to Google. Google also says the Gemini agent can operate as a headless agent inside third-party applications, meaning it can run in the background without a conventional visible interface. This broad reach is useful for teams with mixed software environments, but it increases the importance of understanding where an agent is active.

Gemini’s cross-platform design matters because work often starts in one application and ends in another. A research request might begin in chat, use documents and spreadsheets for context, and return an output through email or a collaboration tool. A shared agent memory could make that process more continuous, although the same continuity can create confusion if teams do not know which context the agent retains.

Google Workspace users should review existing sharing and access practices before connecting work data to an agent. Employees who already use Gemini controls in Gmail may recognize the broader privacy question: useful AI features depend on what data is made available to them.

What work can the Google Gemini agent do over time?

Google’s Gemini agent can keep cloud-based work running for hours or days after a user closes a laptop, according to Google. The product can also create temporary sub-agents for multi-step work and coordinate parallel or sequential tasks between those agents. The capability could help with long research, coding, document, or media workflows that do not need constant manual input.

Google’s description of temporary sub-agents suggests that a larger task can be divided into smaller assignments. Parallel work means separate sub-agents can handle different parts of a task at the same time, while sequential work means one result can feed into the next stage. The practical limitation is that splitting a task does not guarantee that every intermediate result is correct, complete, or safe to use.

Google’s long-running task model is particularly different from ordinary chat use because the work can continue after the person who started it has stepped away. Teams should define an expected finish condition, an output location, and a human owner before allowing an agent to begin a task that can run for days.

Why does persistent AI work create a security concern?

Google’s Gemini agent creates a governance concern because a persistent system can retain assigned context and continue work after the initiating user is no longer actively watching it. Google says coworker agents receive access only to context supplied by a user or team, which is an important stated boundary. The remaining risk is that users may provide too much context, connect an unnecessary system, or approve an output without checking it.

Persistent execution changes the normal security review from “What did I type into the chatbot?” to “What can this agent continue doing after I leave?” That distinction matters for documents, email, calendars, code, and other business data where an agent may be able to see relationships between separate sources. The practical response is to require clear task ownership, narrow access, and a review step for any action that affects customers, spending, legal obligations, security controls, or production systems.

Organizations should also treat AI-agent security as part of their broader incident planning. Government attention to AI incident reporting expectations reflects why companies need clear records of what automated systems can access and who is responsible when a workflow fails.

What controls should organizations use before enabling Gemini agents?

Organizations should begin with least-privilege access, defined tasks, human approval points, and audit-friendly ownership for every Gemini agent. Google says Gemini can carry the same memory, skills, and controls across Workspace surfaces, so a decision made at setup can affect more than one application. The safest initial deployment is a limited pilot using non-sensitive information and low-consequence tasks.

1. Define the task outcome before granting access, including what the Gemini agent may read, create, or send.
2. Limit the Gemini agent to the minimum documents, mailboxes, folders, or applications required for that task.
3. Require a human reviewer before the Gemini agent sends external messages, changes records, runs consequential code, or approves spending.
4. Record the agent owner, the connected systems, and the process for disabling access when a project ends.
5. Stop the deployment and contact the organization’s security or IT team if the agent receives unexpected access, produces sensitive output, or begins work outside its defined scope.

Google’s Gemini Enterprise release notes also say administrators can control access to Antigravity’s multi-agent Teamwork feature, which remains in preview. Google’s Gemini Enterprise release notes identify the administrative control point, but preview status means organizations should validate behavior in their own environment before treating the feature as a stable production dependency.

What does Google Gemini Enterprise cost?

Google Gemini Enterprise Business starts at $21 per seat per month and includes a 30-day trial, according to Google. Google says the Business tier supports up to 300 seats and does not require IT setup, while Standard and Plus plans start at $30 per seat per month. Those are starting prices, so organizations should confirm current plan terms and eligibility with Google before purchasing.

Google’s pricing structure matters because a per-seat plan can become materially different in cost as access expands from a small pilot to a full department. A lower entry price may suit a limited test, while broader deployment can require stronger internal policies, administrator oversight, and budget approval. The practical decision should consider the cost of review and governance alongside the subscription price.

Teams should not assume that a low-friction setup removes the need for technical oversight. Google says the Business plan does not require IT setup, but security, legal, and data-governance teams may still need to establish internal rules before employees connect work information to persistent agents.

Who should use the Google Gemini agent first?

Google’s Gemini agent is most suitable first for teams with repeatable knowledge work, clear task owners, and data that can be safely scoped to a defined project. Google says the agent can answer questions, produce media, write and run code, and coordinate multi-step work, making it potentially useful across several work functions. The product is less suitable for unattended use where an incorrect answer or unauthorized action could create immediate harm.

Google’s Gemini agent can be practical for drafting, research organization, internal document preparation, or controlled coding tasks when a person remains responsible for the final output. The product should not be treated as an independent decision-maker for legal, financial, medical, personnel, or security-sensitive actions. Human review remains necessary because persistent execution does not establish accuracy, judgment, or accountability.

For most organizations, a small pilot provides more useful evidence than an immediate companywide rollout. A pilot can identify which data sources employees actually need, where output review breaks down, and whether the agent’s continuing work fits existing approval processes.

FAQ

Can the Google Gemini agent get its own email address?

Yes, Google says persistent Gemini coworker agents can receive an @agents.company.com email address, along with an identity and storage. Organizations should assign that identity only after defining the work context and permissions the agent needs.

Can the Google Gemini agent keep working after I close my laptop?

Yes, Google says the Gemini agent can continue cloud-based work for hours or days after a user closes a laptop. Users should set clear task boundaries and review the completed output before relying on it.

Can Gemini agents use Microsoft 365 and Slack?

Yes, Google says the Gemini agent is accessible through Microsoft 365 and Slack, as well as Google Workspace, web, mobile, desktop, and command line tools. Availability in a specific organization can depend on the administrator’s setup and connected systems.

Does Google Gemini Enterprise include a free trial?

Yes, Google says Gemini Enterprise Business includes a 30-day trial and starts at $21 per seat per month. Organizations should confirm current US pricing and plan terms before beginning a deployment.

Should a Gemini agent have access to all company files?

No, a Gemini agent should receive only the files, systems, and context needed for its assigned task. Limiting access reduces the chance that a persistent agent can expose or use information outside its intended workflow.