Google turns Gemini into a single agent for work, starting with businesses
At Gemini at Work on October 8, 2026, Google Cloud CEO Thomas Kurian introduced the Gemini agent: one agent meant to field questions, carry out knowledge work, produce images and media, and write and execute code. It can get a Workspace account of its own and talks to MCP servers. TechCrunch reports that businesses get it first and consumers later.
The chat box now takes delivery of whole jobs
Google Cloud's blog published the announcement on October 8, 2026, adapted from Kurian's keynote at Gemini at Work 2026. Google pitches the Gemini agent as one universal agent for work, covering everything from quick answers and knowledge work to media creation and coding. Kurian summed up the shift in one line: "Work now starts in the prompt window."
According to TechCrunch, Kurian said the agent can be handed "objectives, not just instructions." In practice the user describes an outcome instead of scripting each step, and the agent plans the work, picks up skills and tools, and reaches into internal company systems. Google says Gemini can spin up temporary sub-agents for multi-step jobs, each built for that one job and carrying its own identity. Such work can stretch over hours or days, and because the agent lives in the cloud it keeps going after the laptop is closed. TechCrunch adds that users can follow the agent's reasoning, the tasks it hands to sub-agents and its progress from a tasks inbox view.
A team member with its own email, calendar and Drive
Google lists the places the agent can be reached: the web, iOS and Android, Windows and Mac desktops, the command line, Google Workspace, Microsoft 365 and Slack. The connector list is long too. It covers collaboration tools such as Confluence, Microsoft Office, Teams, Slack and Workspace, developer tools such as Git and Jira, Salesforce and ServiceNow, and databases including BigQuery, Databricks, Postgres and Snowflake. On top of that, the agent can work with any Model Context Protocol (MCP) server, whether it sits on the company network or outside it.
The most striking piece is the coworker agent. Google says it gets a Workspace account of its own, with an email address, a calendar, a Drive and a listing in the company directory. Colleagues pull it into a Chat space or @mention it when they need it; it acts under its own identity rather than the user's and sees only what has been shared with it. TechCrunch notes that every step it takes lands in an audit trail credited to the agent, not to a person. The model is treated as a separate choice: Google says it currently routes work between its Gemini family and Anthropic's Claude models, with other private and open models to follow. Industry editions for financial services and legal are in preview, with government, healthcare and retail next in line.
Cost control and oversight now ship with the product
Google is positioning the agent not as a separate app but as a colleague working inside a company's email, documents and business systems, and MCP support favours businesses that can open their own systems to it. Cost sits at the centre of the pitch as well. Google says per-token prices are down 98% since 2024, while enterprise AI usage has grown sharply. Its answer has three parts: matching the model to the job, smart routing that sorts workloads automatically, and real-time spend caps. A cap is set per project in the Cloud Billing Console; once it is hit, that project's agent stops and can be restarted with one click in the console. Because tracking runs per project, costs can be charged back to individual departments.
On scale, TechCrunch reports Sundar Pichai saying Gemini has more than 1 billion monthly active users. Google adds that Gemini Enterprise is in use at nearly 90% of the Fortune 100, and that nearly 80% of Google Cloud customers use its AI products. TechCrunch lists On, Shopify and PayPal among the agent's early testers. Customer results are presented as company-reported figures: per Google, Ulta Beauty's shopping assistant built with Gemini Enterprise tripled digital sales conversion, and at Wesfarmers, shopping agents lifted conversion for Kmart and Officeworks by up to three times.
The sources say nothing about Türkiye, so this section is commentary
Neither the Google Cloud announcement nor the TechCrunch report mentions Türkiye. What follows is therefore not drawn from the sources; it is UNALSOFT's own commentary. The open questions, one by one: a general availability date and the list of countries, when the agent can be used in Türkiye, pricing for the agent itself, and a date for the consumer version. For consumers, TechCrunch only says later.
Our view: for a business in Türkiye, preparation does not hinge on a launch date. An agent is only as useful as the systems opened to it and the approvals around it. Which of your data is reachable through an API or an MCP server, which tasks need a human sign-off and which project the spend should be booked to can all be mapped today. An agent that works under its own identity and logs its actions to an audit trail may make access rights easier to define from a KVKK perspective, but that does not replace a local legal assessment. We covered the original Gemini Enterprise announcement on July 18.
The UNALSOFT view
We read this less as a model race and more as an integration question. The agent Google describes only delivers work once it is wired into a company's tools, data and approval flow, and MCP support is what makes that wiring standard. Controls such as spend caps and model choice could make agent projects more predictable for SME budgets. The practical order is to define the systems, permissions and budget limits first and pick a platform second. In our agentic AI service we start from exactly that map. Because the sources carry nothing about Türkiye, this article does not suggest a local availability date.
Are your company's systems ready for an agent?
In a short conversation we can work out which of your data and tools could be opened to an agent, and which jobs should run under a budget cap.