Google upgrades Gemini with agentic AI push aimed first at businesses

Google is turning Gemini into an agent that can do work, not just answer questions, and it is launching the new capability first for enterprise customers. The move matters because Google says Gemini already has more than 1 billion monthly active users and is widely used in business, giving the company a large installed base for the next phase of AI competition.

Announced at a Google Cloud event on Thursday, the new unified agent is designed to take objectives, coordinate tasks, connect to workplace systems and produce an audit trail of its actions, signaling a broader shift in AI from chat to execution.

What Google announced

Google said it is bringing Gemini into what it described as the agentic era: a mode in which the AI does more than respond to prompts and instead carries out tasks on behalf of users through a single interface. The company’s goal is to make Gemini capable of handling multi-step work that previously required a human to move between apps, documents, calendars and internal systems.

In practical terms, the agent can receive a business objective, break it into smaller steps, choose tools, and carry out work across connected services. Google positioned the release as a major extension of Gemini Enterprise, its workplace AI offering, while making clear that consumer access will come later.

Why Google is starting with businesses

Google is prioritizing enterprises because business customers are already using Gemini at scale and because enterprise environments present the biggest technical and security hurdles. Sundar Pichai said the company wants to solve the more difficult issues around security, scale and performance before bringing the agent more broadly to consumers.

That strategy reflects a basic reality of the current AI market: the most advanced agent features can be powerful, but they also raise concerns about access control, error handling, accountability and data governance. By rolling out first to companies, Google can test the system in controlled environments where workflows, permissions and compliance requirements are better defined.

Google Cloud CEO Thomas Kurian said the new system is built to take “objectives, not just instructions,” highlighting that the agent is meant to plan and execute work rather than simply wait for step-by-step commands.

How large is Gemini’s existing footprint?

Gemini’s scale gives Google a strong starting point. Pichai said Gemini now has more than 1 billion monthly active users, and he added that nearly 90% of Fortune 100 companies use Gemini Enterprise in the workplace. That combination of consumer reach and corporate adoption could help Google move faster than rivals that are still building distribution.

The company is banking on the idea that a product already embedded in everyday work and consumer habits can be upgraded into a general-purpose agent without asking users to adopt an entirely new platform.

How the new Gemini agent works

The core idea behind the new release is delegation. Instead of merely responding to prompts, Gemini can be given an objective and then decide how to fulfill it using skills, tools and connected data sources. Google says the agent can plan its own workflow, select the right model by default, and interact with internal systems as needed.

Users can also override the default model selection. The first third-party option will be Anthropic’s Claude models, and Google said the picker will eventually expand to include open-source models and other private models as well.

What tools and systems can it connect to?

Google says the agent can connect to a wide range of enterprise platforms. Supported systems include Google Workspace, Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Postgres and Snowflake, among others. It can also work securely with Model Context Protocol servers whether they sit inside or outside a company’s network.

That breadth matters because agentic AI becomes useful only when it can operate across the software stack where employees already work. Rather than being confined to a chatbot window, Gemini is being positioned as an orchestration layer for corporate knowledge and action.

• Accesses files, folders and other work artifacts
• Uses special skills and tools tied to specific workstreams
• Connects to workplace software and data warehouses
• Works through secure MCP connections
• Produces an audit trail for accountability

Why the “tasks inbox” matters

The new tasks inbox is one of the most important parts of the launch because it makes the AI’s work visible. Google says users will be able to see the system’s reasoning, delegated sub-tasks, special skills being loaded, code execution and progress updates from a central interface.

That transparency is designed to help businesses trust an AI that is acting autonomously. When software is allowed to draft responses, move data or trigger actions across systems, companies need a way to understand what happened, why it happened and who or what is responsible if something goes wrong.

In other words, the inbox is not just a convenience feature. It is a governance feature.

How Google is making Gemini act like a coworker

Google is giving the agent its own Workspace account, which means it behaves much more like a digital employee than a conventional assistant. The company says Gemini will have its own email address and its own working context, allowing it to understand team structures, time zones, approval chains and calendar availability.

Users will be able to summon the agent by tagging it, emailing it, sharing files with it or adding it to a group chat. As it completes work, it will leave an audit trail under the agent’s identity rather than under a human employee’s name.

Google says this identity-based setup is meant to make the agent easier to track inside normal workplace processes, especially when multiple people and approval layers are involved.

What does this mean for workplace workflows?

It means companies can assign tasks to Gemini in much the same way they would assign work to a teammate. The difference is that Gemini can operate across documents, chat apps, analytics tools and project-management systems without switching context the way a person would.

That could be useful for recurring office work such as preparing briefs, coordinating meetings, pulling data, generating code or routing documents through approval steps. It also suggests that Google sees agentic AI not as a novelty but as a layer that may sit inside the daily mechanics of enterprise software.

Which companies are testing it already?

Google said early testers include On, Shopify and PayPal, three companies with very different operational needs but a common interest in automation, workflow efficiency and cross-platform collaboration. The company also named several Gemini Enterprise customers, including BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty and Wesfarmers.

Those names suggest that Google is aiming the product at a broad mix of sectors, from finance and healthcare to retail, utilities and consumer brands. The pitch is not limited to one vertical: it is meant to be general enough to fit most knowledge-work environments.

How Google is addressing cost and control

Agentic AI can become expensive quickly, especially when it involves multiple model calls, routing decisions and long-running work. Google said it is introducing flexible spending controls to help businesses manage those costs, including multi-model orchestration, smart routing and real-time spend caps.

Those controls are likely to matter to enterprise customers that want the power of advanced AI without losing budget discipline. As companies scale AI usage across departments, finance teams increasingly want better visibility into how much each workflow costs and which models are doing the work.

Why cost controls are strategically important

Cost transparency can determine whether agentic AI becomes a broad workplace standard or remains a niche premium tool. If companies cannot predict usage expenses, they may restrict deployment or limit the number of workflows they automate.

Google’s decision to highlight spend management shows that it understands the enterprise AI market is now as much about economics as performance.

How Gemini fits into the broader AI agent race

Google’s announcement arrives at a moment when the tech industry is moving quickly from conversational assistants to autonomous or semi-autonomous agents. Competitors have already begun exploring consumer-facing and messaging-based agents, and OpenAI’s recent Dots launch has intensified the sense that the category is evolving fast.

Google’s response is to emphasize breadth, integration and enterprise readiness. Rather than launching a narrow agent with limited actions, it is wrapping Gemini into a corporate workflow layer that can touch calendars, emails, chat tools, code repositories and data stores.

That approach could be especially dangerous for rivals if it works. Google already has distribution through Search, Android, Workspace and Cloud, giving it multiple routes into daily user behavior. If Gemini becomes the default agent inside those ecosystems, Google may be able to normalize agentic computing faster than companies starting from scratch.

What makes this release different from a normal chatbot update?

This release is different because it changes the unit of work. A normal chatbot answers questions; Gemini’s new agent is meant to complete objectives. That distinction sounds subtle, but it changes how users interact with the system and how businesses think about AI governance.

Instead of asking, “What is the answer?” users can ask, “What should be done?” The agent then decides how to proceed, which tools to use and which steps require human review.

Examples of likely use cases

Google did not publish a full catalog of scenarios in the announcement, but the systems it connected to point to several likely applications:

1. Coordinating meeting schedules across time zones
2. Drafting project updates from multiple data sources
3. Routing files through approval workflows
4. Pulling analytics from business databases
5. Generating and managing code tasks
6. Supporting customer or partner communication workflows

These are the kinds of repetitive, information-heavy jobs that businesses often want to automate first. They are also the kinds of tasks where traceability and permissioning are essential.

What happens next?

Google has made clear that the first phase is for businesses, but the long-term ambition is broader. Once the company is satisfied with security, scale and performance, the same agentic capabilities could eventually be extended to consumers.

For now, the enterprise launch gives Google a chance to prove that Gemini can function as a reliable workplace operator, not just a smart answer engine. If it succeeds, the company will have turned one of the largest AI products in the market into a platform for execution rather than conversation.

That is a significant shift. The next major battle in AI may not be about who can chat best, but who can safely do the most work.

Timeline of Google’s Gemini agent rollout

For now, the message from Google is clear: Gemini is no longer just something to talk to. It is being built to act.