Google has turned Gemini into a full-fledged AI agent for businesses, one that can plan its own work, carry out multi-step tasks and operate across the software stack that companies already run. Revealed at a Google Cloud event on October 8, 2026, the new agent goes beyond answering prompts: it decides how to reach an objective, hands pieces of the job to subagents and picks which AI model should tackle each step. The launch puts Google squarely in the race to own the enterprise agentic AI layer.
From Chatbot to Coworker
The central idea behind the announcement is a shift from instruction-following to goal-seeking. Rather than waiting for a user to spell out each action, the Gemini agent is designed to plan work and accomplish objectives on its own, according to TechCrunch's report from the event.
To make that practical inside real companies, Google has given the agent unusually broad reach. It connects to:
- Productivity suites: Google Workspace and, notably, Microsoft 365
- Collaboration and engineering tools: Slack, Jira, Confluence and Git
- Data platforms: BigQuery, Databricks, Postgres and Snowflake
The agent also supports Model Context Protocol (MCP) servers, the open standard that has become the default way to plug AI agents into external tools and data. That means organisations can wire in internal systems without waiting for Google to build a native connector.
An Agent With Its Own Identity
One of the more striking design choices is that each agent gets its own Google Workspace account, complete with an email address and an audit trail. In practice, that treats the agent less like a feature and more like a digital employee: it can be granted permissions, its actions can be logged, and administrators can review what it did and when.
That identity model speaks directly to the governance worries that have dogged agentic AI this year. Enterprises have been reluctant to let autonomous software act inside their systems without clear accountability, and a dedicated account with an audit history gives security and compliance teams something familiar to manage.
The agent is available across iOS, Android, Windows and Mac, as well as from the command line and through integration platforms, so it can follow employees between devices and into developer workflows.
Multi-Model by Design
Perhaps the most strategically interesting detail is that Google is not locking the agent to its own models. Gemini models are the default, but the agent can also call third-party models, starting with Anthropic's Claude. Google says it intends to expand support to open-source and private models over time.
The agent can also delegate tasks to subagents and automatically select the most appropriate model for each piece of work. Alongside that, Google introduced flexible spending controls, including multi-model orchestration and real-time spend caps, although specific pricing was not disclosed.
For buyers, the multi-model approach reduces the fear of betting on a single vendor's model roadmap. For Google, it is a bet that owning the orchestration layer, and the customer relationship, matters more than which model answers any given request.
Scale and Early Customers
Google is launching from a position of reach. According to the company, Gemini has more than 1 billion monthly active users, and nearly 90% of Fortune 100 businesses use Gemini Enterprise.
Early testers of the new agent include On, Shopify and PayPal. Enterprise customers named alongside the launch include BNP Paribas, Bradesco, Merck, Orange Spain, Santee Cooper, SOMPO, Ulta Beauty and Wesfarmers. The agent is rolling out to businesses first, with a consumer version to follow later.
Why It Matters
The announcement lands in a crowded week for enterprise agents. Microsoft has added "Hooks" to Copilot Studio to make agent workflows more predictable, SAP is expanding Joule into an agentic work layer, and Oracle has embedded an agent orchestration layer directly into its ERP suite. Every major platform vendor now wants to be the place where a company's AI agents live.
Google's pitch differs in three ways:
- Cross-vendor reach: connecting to Microsoft 365 and Slack signals Google wants to orchestrate work even in companies that are not all-in on Workspace.
- Accountable identity: giving agents their own accounts and audit trails tackles the trust problem head-on.
- Model neutrality: supporting Claude from day one positions Gemini as a hub rather than a walled garden.
The hard part, as industry surveys keep showing, is moving from pilots to production. Many large companies are experimenting with agents, but far fewer have deployed them at scale, and analysts expect a meaningful share of agentic projects to be abandoned. Google's emphasis on identity, auditability and spend caps is a direct attempt to remove the objections that keep agents stuck in the lab.
The real contest in agentic AI is no longer about who has the smartest model, but about who controls the layer where agents are given permissions, budgets and accountability.
If Google can make that layer feel safe enough for regulated industries such as banking and pharmaceuticals, both well represented in its launch customer list, Gemini could become the default operating environment for corporate AI agents.
