Google just made its boldest move yet in the race to sell AI that works for you. At its Gemini at Work 2026 event on October 8, Google Cloud CEO Thomas Kurian unveiled the Gemini agent: a single, universal agent for work that lives inside the apps your company already uses and keeps working even after you close your laptop.
The pitch is simple. Instead of opening a chatbot, pasting in a document, and hoping for the best, you give the agent an objective and it plans the work itself. It pulls context from your business systems, uses tools and custom skills, splits long jobs across temporary sub-agents, and hands back finished work inside Gmail, Docs, Sheets, or your developer environment. “You give it objectives, not just instructions,” Kurian said on stage in Mountain View.
A single prompt box across Gmail, Docs, Slack, and even Microsoft 365
The agent runs in the cloud and holds your business context across every surface it touches. Inside Workspace, that means Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar. It remembers who you normally meet with, so you can ask for a meeting with your usual team without listing every attendee. It can also reach outside Google’s walls: Slack, Microsoft 365 (including Teams), the command line, and developer tools like Git and Jira.
That breadth matters because nobody’s actual workday fits inside one app. Google’s example flow says it well: ask the agent to research a market, build the financial model in Sheets, and draft the presentation from the findings, all from one prompt. The same memory carries over between devices, so work started on your phone continues on your desktop.
The standout feature: it picks between Gemini and Claude
Here is where Google zigged where everyone else zagged. The Gemini agent is model-agnostic. For each task, it picks the best model for the job, and at launch that choice set includes Google’s own Gemini models and Anthropic’s Claude models. Google says more models will be added later.
This is a genuine break from the usual playbook. OpenAI’s Dots agents run on OpenAI models, Microsoft’s Copilot stack runs on its own partnership models, and Anthropic’s offerings run Claude. Google is effectively telling enterprise buyers they do not have to bet their whole agentic workforce on one lab’s models. The agent also ships with built-in cost controls, a quiet but important detail given that nobody has yet announced pricing for the Gemini agent itself. As our October 2026 AI price list shows, model pricing has been falling fast, and routing the right task to the right model is how companies actually capture those savings.
Coworker agents get their own email address
Beyond the personal-assistant mode, Google is pushing a second idea: “coworker agents” that act as team members. A coworker agent gets its own Workspace account, email address, and calendar, and can be delegated long-running work. The catch, and it is a good one, is that it only sees the information the team explicitly shares with it.
Industry-tuned versions are coming too. Variants built for financial services and legal work are already in preview, and versions for government, healthcare, and retail are on the roadmap. That follows the Gemini 4 Argon playbook, where Google started with cyber-defense teams rather than a general release.
Google’s scale argument: 500 trillion-token customers and 80% adoption
Kurian’s keynote came with numbers aimed squarely at CIOs. Nearly 500 Google Cloud customers each processed more than one trillion tokens in the past year. Nearly 80% of all Google Cloud customers now use its AI products, and nearly 90% of the Fortune 100 use Gemini Enterprise.
The early customer stories were specific, which is rarer than it should be in these announcements. Sportswear brand On tested the agent’s dynamic model-selection capability. Shopify is blending frontier models across its merchant base. PayPal is routing 10 million multi-model requests every week. Brazilian bank Bradesco cut a document review workflow from one hour to five minutes, and telecom operator Orange Spain has deployed more than 1,000 custom Gemini Enterprise agents.
Even so, the usual caveats apply. The agent is in private preview with a select set of customers, with wider rollout planned for select Workspace Business and Enterprise plans. No pricing and no general availability date have been announced. And like every agent launch this year, the real test is not the demo but the audit trail: whether IT can see what the agent did, why, and what it touched. Google is promising enterprise-grade security, administration, and governance out of the box, but buyers will judge that in pilots, not keynotes.
What this means for the agent race
The timing is not a coincidence. OpenAI shipped always-on Dots agents in September. Meta released Muse, a personal agent that shops, books travel, and sends emails, last month. Every major lab is converging on the same insight: chat was the prototype, and agents doing real work are the product.
Google’s differentiation bet is the workspace itself. Most companies’ data already lives in Gmail, Drive, and Sheets, and Google is the only vendor that can put an agent natively inside those apps rather than bolting it on. If you are already a Workspace shop, the Gemini agent arrives with your data, your permissions, and your IT controls attached. That is a shorter sales cycle than any chatbot ever managed.
The open question is the one nobody on stage answered: what it costs. Until Google names a price, the Gemini agent is the most interesting free demo in enterprise software. Watch for the Business and Enterprise plan announcements, because that is when the race stops being about keynotes and starts being about procurement.
