Gemini Agent: Google's AI coworker that tackles tasks on its own

What's new

Google Cloud has unveiled Gemini Agent, a unified AI agent for work tasks. An agent is an AI program that does not just answer a question but gets started solving a task on its own. Here, the user specifies a goal instead of a series of individual instructions and gets finished work back.

From a single interface, the agent can answer questions, carry out knowledge work, create content and media, and write and run code. It can be used through Google Workspace (Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar), Microsoft 365 and Slack, as well as other apps.

The agent is in closed beta, so only select users can try it. According to Constellation Research, general availability is expected in late October or early November 2026. The sources do not fully agree on the timeline, however. Early users include On, Shopify and PayPal.

What's smart

The agent plans the work itself using custom skills and tools. It can even create sub-agents with their own roles and identities, a bit like a manager delegating tasks to a small team.

It can run scheduled tasks and respond to events. It can also keep working in the cloud for hours or days without the user being online. The agent has several kinds of memory: session (what is happening right now), semantic (knowledge), procedural (how things are done) and episodic (what has happened before).

The agent connects to enterprise systems such as Confluence, Jira, Microsoft Office and Teams, ServiceNow, Snowflake, Salesforce, Amazon S3, Azure Data Lake and Databricks Unity.

On the security side, agents get their own identities and role-based permissions. There is also auditing (a trail of what the agent has done), policy enforcement, sandboxing (isolated environments) and spending caps. An Agent Gateway manages networking and policies. Thomas Kurian, CEO of Google Cloud, says, in paraphrase, that it should be connected to your personal workflows, your systems and your enterprise controls.

Cheaper or better?

Pricing has not been disclosed yet. According to Constellation Research, the pricing model will be usage-based with no separate product tiers, but details on the agent's usage costs are still to come. There is no information on monthly pricing for regular users or on price per million tokens.

The agent supports both Google's own Gemini models and Anthropic's Claude. According to IT Brief UK, the models mentioned include Gemini 3.1 Pro, Gemini 3.1 Flash Image, Veo 3.1 Lite and Lyria 3 Pro, as well as Claude Opus, Sonnet and Haiku, including Claude Opus 4.7. Several of the models are in preview. More closed and open models are expected to be added later.

Analyst Holger Mueller of Constellation Research calls the open approach especially attractive, because companies can use the language models they trust within the agent's automation. Kurian himself positions Google as more vertically integrated than other large cloud providers, with its own models and its own chips in one platform. Direct benchmarks against competitors have not been disclosed.

What it's good for

Gemini Agent is aimed at companies that want AI to handle entire workflows across email, documents, spreadsheets and line-of-business systems, rather than only getting help with individual tasks. Because the agent can keep working for hours or days, it suits heavy or recurring tasks that would otherwise eat up time.

Google says that almost 80 percent of Google Cloud's customers use the company's AI products, and that almost 90 percent of Fortune 100 companies use Gemini Enterprise. Orange Spain has more than 1,000 custom agents, and SOMPO has more than 10,000. The figures are Google's own and have not been independently verified.

The sources do not say whether or when Gemini Agent will become available in Denmark and the EU.