Google's Gemini Agent: A Universal Work Agent for Cloud and Workspace

On October 8, 2026, at the Gemini at Work 2026 keynote, Google Cloud CEO Thomas Kurian introduced 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 and one API. The announcement lands in a market where OpenAI and Meta have already shipped their own always-on agents, and it reframes what Google expects knowledge work to look like.

For the past two years, enterprise AI has mostly meant a chat window: ask a question, get an answer, copy it somewhere useful. Google Cloud's Gemini at Work 2026 keynote argued that model is finished. In its place came the Gemini agent — what CEO Thomas Kurian called "your new single, universal agent for work" — announced on October 8, 2026.

Google's framing is "objectives, not instructions." You delegate an outcome — a research brief, a quarterly analysis, a working prototype — and the agent plans the work, uses skills and tools, connects to your systems, and brings back finished work inside the documents, inboxes and developer environments you already use. It can act as a personal assistant or as a team member: a project manager within a team, or an analyst inside a finance department.

What the Gemini agent actually does

Google describes three architectural principles. First, a unified agent: chat, autonomous objective execution and code generation from a single interface — you can assign work, schedule tasks, or have it respond to events. Second, omnipresent access: web, iOS and Android, Windows and Mac desktops, and channels including the command line, Google Workspace, Microsoft 365 and Slack; it can also run headless, with no dedicated user interface. Third, persistent execution: the agent runs in the cloud with one set of memories, context and a personalization graph across every device and channel, so work keeps running after you close your laptop.

Reuters, reporting the announcement the same day, confirmed the core claims: the agent plans work, uses tools, connects to business systems and returns finished work in documents, email and developer environments; it works directly inside Gmail, Docs, Sheets and Calendar and can be used through Microsoft 365 and Slack; and it picks the best model for each task, currently running on Google's Gemini models and Anthropic's Claude models, with others to be added later. Alphabet shares were up marginally in early trading.

Inside Workspace: three ways it works

Gemini now works inline in Gmail, Drive, Docs, Slides, Sheets, Chat and Calendar, carrying the same memory, skills and controls it has everywhere else. Google describes three modes. Personal assistance: Gemini arrives briefed from your calendar, team and documents — you can ask it to set up a meeting with "the usual team" of regional event leads without supplying a single name, and it works out who that is, checks calendars and starts the coordination thread. Proactive delegation: Workspace Intelligence recognizes a delegatable task — a manager's email asking for the latest project update as a slide deck — and offers a one-click handoff, applying the same reasoning to your inbox to surface the message that matters most rather than the one that arrived last. Team membership: you describe a role, and Gemini creates a coworker agent with its own Workspace account — email, calendar, Drive and directory presence — that colleagues work with like anyone else, by adding it to a Chat space or @mentioning it.

Sub-agents and coworker agents

For multi-step work, Gemini can dynamically create a roster of temporary, job-specific sub-agents — each with its own identity — and coordinate parallel and sequential steps that can run for hours or days. Coworker agents go further: persistent, defined roles with their own @agents.company.com email addresses and persistent storage, and access only to the context you or your team provide. A coworker agent acts under its own identity rather than yours, and access follows the sharing and membership your team already uses.

Model choice, cost controls and the competitive race

The agent and the model are now separate choices. Gemini orchestrates each job across Google's Gemini family and Anthropic's Claude models today — other leading private and open models are promised later — selecting per task via Smart Routing. Google's argument: matching the model to the work raises accuracy on hard jobs and lowers cost on simple ones, and since the leading model changes every few months, keeping the choice open means your context, skills and data stay put when it does. Real-time spend caps and multi-model orchestration round out the cost controls. Google says PayPal routes 10 million multi-model requests every week and Shopify blends frontier models for millions of merchants.

The announcement is also a competitive move. Reuters notes it follows OpenAI's September launch of always-on agents called dots that chase user goals across apps on their own, and Meta's Muse, a personal agent that can shop, book travel, send emails and make payments. The enterprise agent race now has three major platforms shipping persistent, goal-directed workers within a month.

Skills, tools and memory

Three layers give the agent business context. Tools: connections to Confluence, Microsoft Office, Teams, Slack, Salesforce, ServiceNow, Git, Jira, BigQuery, Databricks, Postgres and Snowflake, plus desktop files and any Model Context Protocol (MCP) server inside or outside the company network, with an enterprise tools registry so teams can build and publish tools for the rest of the company. Skills: reusable sets of instructions, knowledge or workflows stored as modular prompts — a global library, a shared company registry, and personal skills — with Gemini choosing the right skills and improving consistency from each execution. Memory: four kinds — session memory for the task at hand, semantic memory (a structured knowledge base built from documents and conversations), procedural memory for how a job gets done, including skills it writes for itself, and episodic memory of everything it has done before.

Grounding the answers in data

For data work, Gemini integrates with the Knowledge Catalog (map business definitions like "net margin" once, and every agent uses them), Smart Storage (enriching unstructured objects in place — Google notes roughly 90% of enterprise data is unstructured), and a "borderless Lakehouse" that queries Amazon S3 and Azure Data Lake with no variable egress fees and federates open Iceberg tables across Databricks Unity, Snowflake Horizon and AWS Glue. Bloomberg Media lifted SQL query accuracy by 63% during initial development by grounding its AI in the Knowledge Catalog; its CTO, William Anderson, credited "a trusted institutional context" for the accuracy. Etsy reports data joins up to 60% faster after migrating three petabytes, and Snap cut technical diagnostic troubleshooting from 30 minutes to 30 seconds. These are customer-reported figures, not independent benchmarks.

Industry specializations

Gemini for Financial Services (in preview) automates investment research, credit analysis and risk modeling, drawing on FactSet, LSEG, S&P Global and SEC filings with more than 50 foundational skills; Google says CME Group and Deutsche Bank are early users, with reasoning shown through confidence scores, explicit methodologies and full data lineage. Gemini for Legal (in preview) inherits matter-level permissions and ethical walls from NetDocuments and iManage; Cooley is building a confidential-information redaction agent for litigation filings. Government, healthcare and retail versions were announced as coming soon.

The scale Google claims — and the caveats

Google's keynote numbers: nearly 80% of Google Cloud customers now use its AI products, nearly 90% of the Fortune 100 use Gemini Enterprise, and in the last year nearly 500 customers each processed more than one trillion tokens. Customer examples include BNP Paribas deploying Gemini Enterprise inside LLM@CIB for more than 65,000 employees, Orange Spain running over 1,000 custom agents, SOMPO building over 10,000 agents across 34,000 employees, and Wesfarmers' Bunnings saving a customer-reported half-million hours of administrative work. Treat all of these as vendor-sourced claims: none has been independently audited, and ROI figures like Bradesco's "one hour to five minutes" document review are self-reported.

What is not yet known

Google has not published pricing for the Gemini agent or coworker agents, a general-availability timeline for the industry specializations beyond preview, or independent evaluations of agent accuracy on enterprise tasks. The governance surface — identity, policy management, authorization and permission controls, secure sandboxing and network gateways — was announced in outline; how it behaves across complex permission hierarchies is untested by third parties. And the bar is moving fast: dots and Muse are already in the market, and enterprise buyers should expect the model roster, integrations and pricing to shift within months.

Bottom line: if your organization runs on Google Workspace or Google Cloud, the Gemini agent changes the build-versus-buy math for internal agents. Before delegating real work, audit data permissions and map your business definitions in the Knowledge Catalog — agent accuracy depends on institutional context — and set spend caps early, because persistent agents that run for hours or days consume compute whether or not anyone is watching.

Prices and benchmark figures are standard published rates as of October 9, 2026 and can change without notice. Some outbound links may be affiliate links — see our disclosure. Found an error? Tell us and we'll correct it in public.