At Gemini at Work 2026 on October 8, Google Cloud CEO Thomas Kurian introduced the Gemini agent: one agent that answers questions, carries out multi-step work and writes and runs code from a single prompt box. It runs in the cloud, keeps one memory across devices and channels, and picks the model for each job, Anthropic's Claude included. The design is coherent, and the governance controls are the most detailed any vendor has announced this year. Google is also the last of the big four to ship this kind of agent, it has not given a date for general availability, and its cost depends on consumption rates Google has not published.
The name
Google is simply calling it Gemini. In its analyst Q&A, the company said Spark, its consumer agent, has folded into it, and that the agent is the same whether you reach it from Workspace, the desktop or the command line. Google's comparison is Gmail: one product for consumers and businesses, with different controls and licensing behind it.
That choice makes sense. The Gemini app has 950 million monthly active users, according to Alphabet's Q2 results, so employees meet the agent already knowing the brand. Microsoft made the same choice with Copilot: at general availability, Copilot Cowork runs on Anthropic models, with OpenAI's GPT 5.5 available to Frontier customers, all under the Copilot name.
The name already covers several products. Google sold Gemini as Workspace add-ons until it folded AI into Workspace plans in January 2025, and Gemini Enterprise has been the name of its agent platform since it launched last October. The models keep the Gemini name too, and Google says capability labels such as Gemini 4 Argon will separate model names from product names. An agent called Gemini that sends some of its work to Claude depends on that separation sticking. Where Workspace ends and Gemini Enterprise begins is still unclear, and Google's own explanation covers capabilities more than cost. One name works for users. Buyers still need Google to spell out which SKU includes what.
Workspace integration
Inside Gmail, Docs, Sheets, Slides, Drive, Chat and Calendar, Gemini works inline with the same memory, skills and controls it carries elsewhere. Asked to set up a meeting with "the usual regional event leads," it works out who they are from a Chat space and an old thread, checks calendars and starts the email. Workspace Intelligence flags requests the agent can take on, such as a manager asking for an update deck, and offers a one-click handoff.
The more consequential piece is the coworker agent. Describe a role and Gemini creates an agent with its own Workspace account: email address, calendar, Drive and a directory entry. Colleagues @mention it in Chat or tag it in a comment, its edits appear under its own name in version history, and it sees only what is shared with it.
That visibility is what regulated customers need. It also raises questions the keynote left alone: who reviews the agent's work, who answers for it when it gets something wrong, and how a team's sense of shared ownership shifts when one of its members is software. Those are management questions, and they will decide how far coworker agents spread inside a company.
On price, Google says there is no incremental seat license for the agent in Workspace or Gemini Enterprise. That extends its January 2025 decision to fold Gemini into Workspace plans, which Google said left a Business Standard customer paying $2 more a month than Workspace without Gemini. Microsoft has taken the other route. As I argued in July in Copilot Seats Are the Bridge for Agents, the Microsoft 365 Copilot seat is best read as an installation base: it is the prerequisite for Copilot Cowork, which Microsoft bills on usage on top of the license. Google's analyst Q&A shows both companies end up in a similar place. When the Gemini agent steps outside Workspace to run cross-system work, it draws on Gemini Enterprise's consumption-based capacity. That work is billed on usage, which is why Google launched project-level spend caps with the agent. Google has yet to say whether a coworker agent with its own account consumes a seat.
Competitor landscape
Microsoft took Copilot Cowork to general availability on June 16. OpenAI launched Frontier on February 5. Anthropic made Claude Cowork generally available on April 9. All four now make the same promise: hand over an outcome, get finished work back, and govern the agent the way you govern an employee.
Anthropic is in an unusual position. Its engine powers Microsoft's Cowork, its models run inside Google's agent, and it sells its own product to the same buyers. Each vendor builds on what it already sells. Microsoft works inside the Microsoft 365 tenant, OpenAI sells an operating layer backed by forward-deployed engineers, and Anthropic starts from the knowledge worker's desktop. Google leads with data and infrastructure: a Knowledge Catalog for shared business definitions, a lakehouse that queries Amazon S3 and Azure Data Lake without variable egress fees, and TPUs it says deliver 80% better price-performance than the previous generation.
Google is also going after Microsoft customers directly. The agent works inside Microsoft 365, Teams and Slack, so a company standardized on Office can buy it without moving. Ryanair, rolling out Workspace and Gemini Enterprise to 35,000 employees under a dual-cloud strategy, is the kind of mixed estate Google is after.
Gemini vs. Microsoft Copilot deployment
Microsoft reported more than 30 million paid Microsoft 365 Copilot seats in its July earnings, up from 15 million in January. Against the more than 450 million paid Microsoft 365 commercial seats Microsoft reported in January, that is under 7%, so even strong growth shows how slowly a per-seat add-on spreads. Microsoft also says Agent 365 had nearly 40 million agents registered two months after launch.
Google reports different metrics. Its last seat figure was 8 million Gemini Enterprise seats in February. Since then it has cited nearly 90% of the Fortune 100, nearly 500 customers processing over a trillion tokens each, and the Department of Defense putting Gemini in front of 3 million personnel. A Fortune 100 company counts whether one team uses Gemini or everyone does, and with Gemini bundled into Workspace there is no add-on seat to count.
Model choice is open, while the memory, context and skills an agent builds up stay in Google's layer, and that is where switching costs will be. The same holds for Microsoft. Whichever agent a company lets learn its work first becomes the hardest one to replace.