Google Cloud launches the Gemini agent: a single universal agent for work — delegate objectives, get back finished work

Google Cloud has unveiled the Gemini agent, a single, universal agent for work announced today at Gemini at Work 2026. It takes objectives instead of instructions, picks the best model for each job — Gemini or Anthropic's Claude — and keeps working for hours or days across Workspace, Microsoft 365, and Slack.

Google Cloud today made its biggest enterprise-AI-agent move of the year. At its Gemini at Work 2026 event, the company announced the Gemini agent, a single, universal agent for work that answers questions, handles knowledge work, creates images and media, and writes and runs code — all from one prompt box. The framing from Google's announcement: work now starts in the prompt window.

The pitch is delegation, not chat. You give the agent objectives instead of instructions; it plans the work, picks up skills and tools, connects to your business systems, and brings back finished work inside the documents, the inbox, and the developer environments you already work in. According to a detailed post from Google Cloud CEO Thomas Kurian, adapted from his keynote address, the agent can function as a personal assistant or as a team member — a project manager for a group, an analyst for a finance department — with the security, administration, and governance enterprise customers require.

It keeps working after you close the laptop

Persistence is the architectural bet. The agent runs in the cloud and keeps a single set of memories, context, and one personalization graph no matter which device or channel you reach it from — web, iOS and Android, Windows and Mac, command line, Google Workspace, Microsoft 365, or Slack — and it can run headless, with no dedicated interface at all. Work that takes hours or days keeps running after you close the laptop, and it is still there when you come back.

Gemini does not work alone. It can spin up a dynamic roster of sub-agents — temporary, job-specific agents, each with its own identity — to tackle multi-step tasks in parallel or in sequence over hours or days. "Coworker agents" get more permanent standing: dedicated identities with their own @agents.company.com emails, persistent storage, and a presence in the company directory, with access only to the context their teams provide. In Chat they show up under their own names; in Docs they can suggest edits and reply in comment threads.

Model choice is a separate decision from the agent

Here is the move rivals should study: Gemini is the agent, and the model underneath is a separate choice. Every job runs on the model that fits best — orchestrated across the Gemini family and Anthropic's Claude models today, with other leading private and open-weight models planned. The logic is economic as much as technical: the best model for a task is not always the largest, so matching the model to the work raises accuracy on hard jobs and lowers cost on simple ones — and when the leading model changes, your context, skills, and data stay put.

Early testers already back the routing story: sportswear brand On tested the dynamic model-selection capability to speed time to market; Shopify blends frontier models across millions of merchants; PayPal routes 10 million multi-model requests every week. On the controls side, Google is shipping multi-model orchestration, Smart Routing, and real-time spend caps, with governance through identity and policy management, authorization and permission controls, secure sandboxing, and network gateways.

The scale numbers behind the launch

Kurian's keynote carried the receipts for the timing: in the last year, nearly 500 Google Cloud customers each processed more than one trillion tokens; nearly 80% of all Google Cloud customers are using its AI products; and nearly 90% of the Fortune 100 use Gemini Enterprise. The claim underneath: enterprises are done experimenting and are now running their businesses on this.

The customer roster is unusually specific. Bradesco, one of Brazil's largest banking groups, cut document review time from one hour to five minutes while reducing risk inconsistencies by 60% and booking over 10% in financial efficiencies. Orange Spain deployed more than 1,000 custom agents across HR, IT, sales, and customer service. Japan's SOMPO built over 10,000 custom AI agents across 34,000 employees. Bunnings, the Australian retailer, saved staff half a million hours of administrative work with an internal agent; Ulta Beauty tripled digital sales conversion with an AI shopping assistant; and South Carolina utility Santee Cooper expects 75% faster modeling speed on a $10 billion grid expansion budget.

Inside the suite, Gemini works inline in Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar, carrying the same memory, skills, and controls everywhere. New data and analytics skills turn plain-language questions into operational insights; industry-specific versions target financial services and legal teams. The plumbing matters too: a tools registry connects to Confluence, Salesforce, ServiceNow, BigQuery, Snowflake, Jira, Git, desktop files, and any MCP server inside or outside the company network — while skills, reusable prompt-based workflows, can be published to a shared company registry and improve with every execution.

The open question

The caveat: the universal agent is in private preview, with wider availability for customers on select Workspace Business and Enterprise plans coming soon, per coverage of the event. Google's bet is that objectives-in, finished-work-out becomes the default interface — and that persistent agents with their own email addresses and memories stop sounding strange. On today's evidence, the strangest part may be how quickly "agent as coworker" is becoming an org-chart line item.