Google Cloud launches Gemini agent to work across enterprise systems

Google Cloud launched Gemini agent, a single universal agent for work that can work across business apps and enterprise systems via chat.

Ahead of its Gemini at Work event, Google Cloud laid out a vision where work starts in a prompt window. Gemini agent will have business context and be able to plan work, use skills and tools and connect to enterprise systems. Gemini agent will route the models, include cost control tools and be governed by enterprise policies.

Gemini agent will have one API and be able to operate by using objectives. In a blog post, Google Cloud CEO Thomas Kurian said you give Gemini agent "objectives, not instructions." "You delegate an outcome and come back to finished work," said Kurian. "For an agent to do that, it has to be connected to your personal workflows, your systems of record and your enterprise controls."

Alphabet CEO Sundar Pichai said during the Gemini at Work event:

"Enterprise adoption is moving faster than most people realize. Nearly 80% of all Google Cloud customers are actively using our AI products, and they are not just using it; they're using it intensely. Nearly 500 enterprise customers have each processed over 1 trillion tokens on our platform in the last year. This kind of scale tells us we are moving from a period of experimentation to fully integrating AI into your core business."

Kurian said Gemini agent being in the cloud gives it one set of context to work from. The key word from Kurian is once. Identities for Gemini agent, policies and budgets are all set. Gemini is the agent that will orchestrate other models underneath.

"This has become the new interface for work, but the window is the easy part. The question is, what sits behind it? Today, Gemini becomes an agent. You give it objectives, not just instructions. It plans the work, uses custom skills and tools, connects to your systems, and brings back something finished," said Kurian. "It is one agent. Gemini answers your questions, does your knowledge work, creates your images, and media writes and runs your code, all in one window, all with one agent. That very same agent also orchestrates processes that can run for hours. Give it any role you want: a personal agent, a teammate, or a specialist carrying the skills your company needs. You can assign tasks directly, or have it start on a schedule, or based on an event."

Early customers of Gemini agent were sportwear brand On, Shopify and PayPal.

• Google Cloud starts shipping TPU systems, becomes chipmaker
• Google Cloud revenue growth in Q2 up 82%, nearly hits $100 billion annual run rate
• Google Cloud launches AI cybersecurity agent CodeMender
• Google Cloud Next 2026: A look at the big themes

Kurian outlined the architecture behind Gemini agent including the following:

• Gemini can answer questions and work autonomously from one interface. It can assign work, schedule tasks and respond to events.
• The agent can be access from any device and channel including Google Workspace, Microsoft 365 and Slack as well as third-party applications.
• Gemini agent can maintain context and personalization across channels.
• Gemini agent can create a roster of sub-agents with identities to carry out work and multi-step tasks. Gemini agents have defined roles and identities along with their own emails.
• Context is maintained in Gemini agent across tools, data and history.
• Multiple models can be used. Today, Gemini agent can route to the Gemini family of models and Anthropic's Claude, but Google Cloud said other models, closed and open will be available in the future.

Google Cloud also outlined the skills, context and tools that feed Gemini agent. Gemini can connect to enterprise systems including Confluence, Microsoft Office and Teams, Workspace, ServiceNow, Snowflake, Salesforce and a host of other systems.

Holger Mueller, an analyst at Constellation Research, said:

"Google has systematically rolled out the Gemini franchise with synergies across the portfolio that make Gemini agent very attractive. What stands out is the open approach - which allows enterprises to use already trusted and purchased LLMs as part of the automation scope of Gemini agent."

In addition, Gemini agent comes with a library of skills with tools to build more. Context is maintained and Gemini onboards as a new hire would. Gemini is grounded by a knowledge catalog that maps business definitions, smart storage and a borderless lakehouse that can query enterprise data systems like Amazon S3, Azure Data Lake, Databricks Unity, Snowflake Polaris and more without egress fees.

Within Google Workspace, Gemini agent can work inside of Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar. Gemini can work inside Workspace to offer personal assistance, proactive delegation and act as a coworker or teammate.

Google Cloud also noted that Gemini is backed by a full set of governance tools including identity, authorization and permissions, auditing and policy management and control. The model orchestration, smart routing and real-time spending caps are designed to control costs.

What's next? Kurian said it will be customizing Gemini and its agentic capabilities for industries. "We're specializing the Gemini agent for individual industries with the skills, tools, and knowledge each one needs. Gemini is now in preview for financial services and legal, and coming soon to government, healthcare, and retail," he said.

What you need to know

Here’s a look at the themes and takeaways that stuck out across the Gemini at Work event and an industry analyst session that followed.

• Google is moving Gemini from a chat assistant to a universal work agent that's nested in Gemini Enterprise. In an industry analyst Q&A following the event, executives noted that Gemini agent is in part an answer to personal AI agents that are already presenting governance issues.
• Gemini has layered a host of cost controls and model choices that admins can use. These cost control features including caps on certain types of works as well as model limitations are going to be a big differentiator for enterprises. Google described distinct agent identities, permissions, audit trails, policy enforcement, sandboxing, spend caps, and model routing intended to control cost.
• Model routing is seen as a primary way to control costs. Enterprises can decide to put model routing on autopilot to control costs or limit usage of expensive models. Admins can decide on models, set project-level caps or use individual caps in the coming weeks.
• On the visibility front, Gemini agent's most valuable feature will be budget visibility on agent usage so departments can manage their own investments and returns.
• Consumption based pricing will be the norm for Gemini agent. Google Cloud isn't doing an extra SKU or add-on. The pricing model stays the same and you'll have to gauge what the return on AI will be. In Workplace, current plans will continue and the exact consumption model for Gemini agent usage will come at a later day.
• General availability for Gemini agent, currently in private beta, is expected around the end of October or early November.
• Google Cloud sees its cost control as a key pillar to determining returns. However, ROI can be measured through “speed of innovation” and “additional productivity,” and described the broader accounting lens as “total cost of AI.”
• Industry-specific Gemini editions will be key, but the big question will be the features that win over enterprises. Google Cloud will offer financial services and legal versions for Gemini agent with government, healthcare and retail on deck. Watch what industry specific workflows and processes are baked in.