Google unveils universal Gemini agent to control enterprise AI infrastructure

Google announced a new universal agent and API within its Gemini platform during its Gemini at Work 2026 event this week. The company designed the tool to function as a central orchestrator for enterprise workflows, allowing users to initiate complex tasks from a simple prompt box. This move signals a shift toward controlling the underlying infrastructure layer for artificial intelligence rather than just releasing standalone applications.

According to Computerworld, the agent connects to various skills, tools, and enterprise systems. It can answer questions, generate media, write and execute code, and deliver finished work through documents, inboxes, or developer environments. The system is capable of spinning up sub-agents to assist with specific jobs, enabling a more distributed approach to task completion.

Strategic Positioning and Capabilities

Independent tech analyst Carmi Levy stated that Google is attempting to establish itself as the gatekeeper of the new enterprise operating system. He noted that the company is not merely competing on model quality but is positioning its platform to power whatever underlying model makes the most sense for the user. This approach aims to embed Google deeply into the operational fabric of large organizations.

Google Cloud CEO Thomas Kurian said during his keynote that work now starts in the prompt window. He emphasized that users should provide objectives rather than detailed instructions. The agent operates autonomously from a single interface and can act as a personal assistant or team member, similar to a project manager or analyst.

Access and Governance

The Gemini agent provides omnipresent access, allowing it to function on nearly any device, including iOS and Android phones, as well as Windows and Mac desktops. Users can access the agent through command line interfaces, Google Workspace, Microsoft 365, Slack channels, or integrated third-party applications. It also operates as a headless agent without a dedicated user interface when needed.

Coworker agents can perform tasks across multiple days, sessions, and shifting responsibilities. They possess dedicated identities, including their own @agents.company.com email addresses and persistent storage. However, Google stated that these agents only have access to specifically defined data and channels. Enterprise-dictated security, governance, and cost controls are built directly into the system.

What should you do?

Organizations should evaluate how this centralized orchestration model aligns with their existing security and governance frameworks. Review current access controls to ensure they can manage agents with persistent storage and email identities. Monitor vendor announcements for detailed documentation on how sub-agents handle data across different channels and devices.

What to do and how to stay safe: Google

• Audit your current AI policies to determine if they cover agents with persistent storage and email identities.
• Verify that your security controls can restrict agent access to only specifically defined data and channels.
• Review integration capabilities with your existing tools like Slack, Microsoft 365, or command line interfaces.
• Assess how multi-step tasks running for hours or days fit into your operational monitoring and cost controls.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

What is Google's universal Gemini agent?

It is a single agent and API unveiled at the Gemini at Work 2026 event that orchestrates enterprise tasks across various systems and devices.

How does the agent handle complex tasks?

The agent can spin up temporary or job-specific sub-agents to coordinate workflows involving parallel and sequential steps that may run for hours or days.

Is the agent accessible on mobile devices?

Yes, Google states the agent has omnipresent access and works on iOS and Android phones, as well as Windows and Mac desktops.