Google Cloud has introduced the Gemini agent, a universal enterprise agent designed to complete objectives across knowledge work, content creation, coding, and business systems.
Unlike a conventional assistant that waits for individual prompts, Gemini can schedule work, respond to events, use tools, write and run code, and continue operating in the cloud for hours or days. It carries the same context and memory across Workspace, Microsoft 365, Slack, desktop, mobile, command-line, and headless integrations.
Its most notable capability is multi-agent orchestration. Gemini can create temporary specialist subagents—each with its own identity—and coordinate parallel or sequential workflows. Organizations can also deploy persistent coworker agents with their own company email addresses, calendars, storage, identities, and restricted access.
Google is separating the agent from the model underneath it. The system can route tasks between Gemini models and Anthropic’s Claude models, with additional private and open models planned. Reusable skills, MCP-compatible tools, audit trails, least-privilege permissions, secure sandboxes, policy gateways, and real-time spending controls complete the enterprise-focused architecture.
Google has not published straightforward self-service pricing or a precise broad-availability schedule for the new experience. Independent reporting says the rollout will initially focus on businesses before expanding to consumers.
Why it matters
For builders, the important shift is from selecting one model to designing a durable agent runtime. Context, memory, tools, permissions, identities, and workflows can remain stable while the underlying model changes.
Gemini also raises the standard for enterprise agents: persistent execution, delegated subagents, explicit agent identities, complete audit trails, and enforceable policy boundaries are becoming core platform capabilities rather than optional integrations.
The architecture extends the persistent-coworker pattern seen in Grok Team Bots, but adds model choice, temporary specialist subagents, company-managed identities, and a broader governance perimeter. Teams evaluating it should still test permission boundaries, failure recovery, cost controls, and human approval paths against their own highest-impact workflows.