Google Debuts Gemini, an All-Purpose Agent for Work

Google wants working with AI agents to be as intuitive as Google search was when it launched.

Google Cloud CEO Thomas Kurian introduced the Gemini agent at the Gemini at Work event Oct. 8, stating "Work now starts in the prompt window." Google describes it as a "universal" agent for work, combining the chat, the autonomous worker and the coding agent into one window. 

What the Gemini Agent Does

Google says you give Gemini an objective, not instructions, and it will take the necessary steps to accomplish the work. One example the company shared: ask Gemini to set up a meeting with “the usual team of regional event leads” and it works out who they are from your chat space and past email threads, checks their calendars and starts an email thread to coordinate a time. Or ask it to research market trends, build a financial model in Sheets and turn both into a slide deck, without re-explaining the project at each step.

The agent runs in Google’s cloud, so it retains a persistent memory regardless of how you access it. You can reach Gemini via the web, Android or iOS phones, desktop or a command line. It also works inside Google Workspace, Microsoft 365 and Slack.

For big jobs, Gemini creates and coordinates temporary helper agents, each responsible for one task and each given an identity for auditing purposes.

Gemini can also act as a “coworker” agent with a standing role in a team, such as event coordinator. When used as a coworker, the agent gets its own gmail address, calendar and file storage, and teammates can add it to a chat or tag it like a colleague. It shows up under its own name in a document’s edit history and sees only what members of the team share with it.

Gemini Comes With Limited (for Now) Model Choice

The Gemini agent is separate from the AI model underneath it. The choice is currently limited to Google’s models and Anthropic’s Claude models, but the announcement stated other models will follow. A person can choose their preferred model, or leave it to Gemini to choose — it selects a model based on the job to provide accuracy and keep costs down. Google says this allows companies to switch models later without losing their data or settings.

The Tools Gemini Agents Can Work With 

Google ships Gemini with a number of ready-made skills and four kinds of memory. Organizations can choose which enterprise tools the agent can access or add other connections via MCP.

Part	What it is	What Google says it includes
Tools	The connections that let Gemini work inside the software a company already uses	Slack, Teams, Microsoft Office, Workspace and Confluence for collaboration. Git and Jira for software development. Salesforce and ServiceNow for business operations. BigQuery, Databricks, Postgres and Snowflake for data. Files on an employee's own desktop.
Outside connections	A way to link Gemini to other software via Model Context Protocol	Works with any server using MCP, inside or outside the company network
Shared tool library	A company-wide catalog of tools	Teams can build tools and publish them for the rest of the company
Skills	Saved instructions that teach Gemini how to complete a multi-step task	Ships with a library of ready-made skills
Shared skills library	A company-wide catalog of skills	Departments can publish their own skills, and individuals can build personal ones
Task memory	What it needs to finish the job in front of it	Holds across tasks that run for days
Knowledge memory	Facts it builds up over time	Drawn from documents it reads, people it talks to and other agents it works with
How-to memory	Knowledge of how a job gets done	Includes skills Gemini writes for itself
History memory	A record of what it has already done	Covers every earlier task

How Google Keeps Gemini in Check 

Google says any company deploying agents should be able to answer four questions:

• Who is the agent?

• What is it allowed to do?

• What did it do?

• What should it never touch?

Every Gemini agent gets its own identity and limited access, based on the role. Every time an agent takes an action, it is logged under the agent’s name, not a person's. When Gemini connects to another company’s software, it signs in as itself, using the same standard login system behind “Sign in with Google.” 

Agents run in a walled-off space, and all of their traffic passes through a checkpoint, called the Agent Gateway, which Google calls an "AI network firewall" to enforces company rules. A company writes a rule once, such as “agents may not open documents marked Need to Know,” and it applies to every agent.

What It Costs to Run

Google acknowledged that budgets were under a strain due to rising AI use, even as per-token prices go down. It offered three ways to contain costs:

• Automatic routing sends simple work to cheaper models.

• Companies can set a hard spending cap per project. When it’s reached, the agent pauses until someone clicks to resume.

• Spending is tracked per project, so costs can be billed back to departments.

A limited number of companies already have access to Gemini. A Google spokesperson told Venture Beat Gemini will be built directly into Gemini Enterprise and will be available "at no additional charge for customers where Gemini Enterprise is available, including through existing Google Workspace business and enterprise subscriptions."

What Gemini Suggests About Google’s Strategy

Google is coming late to the agent party after spending much of 2026 chasing Microsoft, OpenAI and Anthropic. Gemini 4 Argon, its newest flagship model, improved its standing, but analysts say the real test comes when businesses use it widely in production. Google's choice to make Gemini an agent that sits on top of any model gives it a second way in the door. If Google can’t always have the best model, it can still be where the work gets done. During the company's Q2 2026 earnings call in July, CEO Sundar Pichai said nearly 90% of the Fortune 100 use Gemini Enterprise.

As independent analyst Carmi Levy told Computerworld, Google is trying to become the gatekeeper of the new enterprise operating system, powered by whatever model makes the most sense. Aside from Gemini's model-agnostic approach and and integration with enterprise systems, its main selling points are controls: agent identities, company-wide rules and spending caps. Pichai made a similar point during the earnings call, telling analysts that the model is just one ingredient in the solutions customers buy.

Have a tip to share with our editorial team? Drop us a line: