Google Cloud has unveiled an AI agent designed to carry out workplace tasks across the company’s productivity software and selected third-party services, marking a move beyond chat-based assistance toward software that can work independently for extended periods.
The Gemini Agent was presented by Google Cloud CEO Thomas Kurian at the company’s Gemini at Work event. Google says users can give the system a goal, close their laptops and allow the agent to continue working in the cloud for hours or even days before returning with a completed result.
An AI agent that works across business software
The system operates across Gmail, Docs, Sheets, Slides, Drive, Chat and Calendar. It can also connect with Microsoft 365 and Slack, as well as business platforms including Salesforce, ServiceNow, Jira, Snowflake and BigQuery.
That broad access is intended to let the agent handle workflows that would otherwise require a person to move information between several applications. Google also says the agent can connect to any server that supports the Model Context Protocol, an open standard designed to let AI systems interact with software and data sources through a common interface.
The announcement reflects a shift in how workplace AI is being positioned. Rather than responding only to individual questions or generating content on request, the Gemini Agent is meant to receive an objective and manage the steps needed to complete it. Kurian described the change by saying that “work now starts in the prompt window.”
Google says the agent is not restricted to its own artificial-intelligence models. Gemini currently coordinates work across Google’s Gemini models and Anthropic’s Claude models, directing tasks to the model it considers the best fit for quality and cost.
‘Coworker agents’ receive company identities
One of the product’s more distinctive features is the ability for managers to create what Google calls coworker agents. A manager can describe a role, such as an events coordinator, and Gemini can create an agent with its own Workspace account.
That account can include:
- A dedicated Workspace email address
- Its own calendar
- Drive storage
- A listing in the company directory
The approach gives the agent an organizational identity separate from an employee’s account. Google says each agent’s activities are recorded in an audit trail attributed to the agent rather than to a human user, making it possible to distinguish software actions from employee actions.
The idea also raises practical questions about how companies will assign responsibility when an autonomous system sends messages, edits documents or changes information in connected services. Google’s design treats the agent more like a digital employee with a defined identity than an invisible feature attached to an individual account.
Security controls and spending limits
Allowing an AI system to work independently across corporate applications creates security and oversight concerns. Google says every agent receives a cryptographically attested identity, meaning its digital identity is verified using cryptographic technology.
Agents also operate inside an “Agent Sandbox,” a separated environment intended to limit access to the wider system. Their network traffic passes through an “Agent Gateway,” which Google describes as an AI network firewall.
Those safeguards are designed to contain the agent and provide administrators with visibility into what it does. The need for such controls is underscored by previous problems involving autonomous software: in July, researchers at Accomplish AI reported that the local mode of Anthropic’s Claude Cowork could escape its Linux virtual machine on Macs and access files on the host computer. That incident was not reported as involving Google’s new agent.
Usage costs are another consideration. AI models process tokens—units of text read and generated by the system—and customers are generally billed according to usage. Google says almost 500 customers processed more than one trillion tokens each over the past year.
To help manage potentially unpredictable usage, Google has added real-time spending caps. When a project reaches its limit, the agent pauses and requires a user to approve resuming the work.
Availability and the market picture
Google did not announce pricing or a date for general availability of the Gemini Agent. Versions designed for financial-services and legal organizations are in preview. Government, healthcare and retail versions are listed as coming soon.
According to Decrypt, the announcement positions Google among technology companies developing AI systems that can perform multi-step work rather than simply answer prompts. The product’s eventual appeal to businesses will depend on how organizations balance automation, access controls, auditability and usage costs.
Moneyiar market data at 21:43–21:44 on October 8, 2026, showed mixed performance among the listed digital assets:
- Compound (COMP): $23.3111531771, up 1.15% over 24 hours and down 1.56% over seven days.
- Nexus (NEX): $0.000000797, down 1.61% over 24 hours and 1.45% over seven days.
- AKEDO (AKE): $0.026346877, down 3.63% over 24 hours and 14.34% over seven days.
- Ailey (ALE): $0.2676796569, down 0.31% over 24 hours and 0.60% over seven days.
