Gemini Agent for Work: What's New, Access, and Alternatives (2026)
The Gemini agent is Google's new single agent for work. Google Cloud announced it on October 8, 2026, at its Gemini at Work event. It answers questions, handles knowledge work, creates images and media, and writes code from one prompt box.
This guide covers what Google announced, how the Gemini agent works inside Google Workspace, its data skills, and who can use it today. It also compares the alternatives. Every detail comes from vendor pages read on October 9, 2026.
What Google announced at Gemini at Work 2026
Google Cloud CEO Thomas Kurian introduced the Gemini agent in his keynote. The Google Cloud blog post adapted from that keynote calls it "your new single, universal agent for work."
The post lists the announcements in one place. They include the agent itself, Gemini working inline in Workspace, and new data and analytics skills. Industry versions for financial services and legal teams, governance controls, and cost controls round out the list.
The post's page data dates it October 8, 2026, and the Google Cloud blog feed carries the same date. The page itself displays October 9, so this guide follows the page data. A companion post for small and midsize businesses carries the same October 8 date.
Google's product page now uses the title "Gemini for Business: The Universal Work Agent." Customer stories on the same blog still refer to Gemini Enterprise, the edition many companies already run.
How the Gemini agent works
The core idea is delegation. In Google's words, "You give it objectives, not instructions." You hand over an outcome, and the agent plans the steps, picks skills and tools, and returns finished work.
Google describes the agent through a set of design principles:
- One agent for chat, tasks, and code. You can ask a question, assign an objective, or have it generate code from the same interface. You can also schedule tasks or have it respond to events.
- Access from many places. The post names web, iOS, Android, Windows, and Mac. It also names the command line, Google Workspace, Microsoft 365, and Slack.
- Work that keeps running. The agent runs in the cloud. The post says "Work that takes hours or days keeps running after you close your laptop."
- Sub-agents for big jobs. The agent can create temporary, job-specific sub-agents to run steps in parallel or in sequence.
- Coworker agents. A coworker agent holds a persistent role on a team, with its own identity and storage.
The result is meant to be a finished piece of work, not a draft in a chat. The post says the agent "brings back something finished" inside the documents, inbox, and developer tools people already use.
Skills, tools, and connections
Three building blocks power the Gemini agent: tools, skills, and context.
Tools connect it to company systems. The post lists Confluence, Microsoft Office, Teams, Slack, and Workspace, plus Git, Jira, Salesforce, and ServiceNow. It also lists databases such as BigQuery, Databricks, Postgres, and Snowflake, and any Model Context Protocol (MCP) server.
Skills teach it how a job gets done. Google defines skills as "reusable sets of instructions, knowledge, or workflows stored as modular prompts." Teams can publish custom skills to a shared company registry, and individuals can build their own.
Model choice and cost controls
The agent and the model are separate choices. The blog post says the agent orchestrates "across our Gemini family of models and Claude models from Anthropic today." The product page lists third-party model choice among features still in early access.
Cost controls are built in. Smart routing sends each job to a model that fits it. Real-time spend caps are set per project in the Cloud Billing Console. When a cap is reached, the post says that project's agent pauses until someone resumes it.
The Gemini agent inside Google Workspace
The Gemini agent also works inline in Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar. Google says it carries the same skills and controls there as everywhere else.
Inside Workspace, the post describes three ways it works:
| Mode | What Google describes |
|---|---|
| Personal assistance | Handles work across apps, such as market research, a financial model in Sheets, and a deck that presents both |
| Proactive delegation | Spots requests it can take on, such as a manager asking for a project update as a slide deck |
| A member of your team | A coworker agent with its own Workspace account, email address, calendar, and Drive |
The coworker agent works under its own identity. The post says it "sees only what you share with it," and its edits appear under its own name in version history.
For deck work, Google's example is concrete. A manager emails asking for the latest project update as a slide deck. Workspace flags the request, and a single click passes it to the agent. Our explainer on Gemini cross-app decks covers Google's earlier deck features.
Data and analytics skills
Google says it is "giving Gemini skills built for specific domains, starting with data." The new skills serve two audiences.
For business users, the post says anyone "can generate reliable, real-time operational reports simply by asking." The agent builds the query and saves it. Once saved, Google says teams can run those reports on demand "without incurring token costs."
For data and ML engineers, the agent generates PySpark code, provides notebooks to edit and test it, trains models, and fixes pipeline issues.
Three Google Cloud services ground those answers. The Knowledge Catalog maps business definitions such as "net margin" once, so every agent uses them. Smart Storage adds context to unstructured files where they sit. The Borderless Lakehouse lets the agent query data in other clouds without copying it.
The output side matters for anyone who builds reports. The post says the Gemini agent "can also generate charts and build dashboards in the tools your analysts already use." These skills are built around Google Cloud data services such as BigQuery and the Knowledge Catalog.
Industry versions for finance and legal
Google is also shaping the agent for specific industries. The post says it is "now in preview for Financial Services and Legal," with government, healthcare, and retail versions coming later.
The financial services version covers investment research, credit analysis, and risk modeling. It draws on data from FactSet, LSEG, S&P Global, and SEC filings. Google says it ships with more than 50 foundational skills.
The legal version inherits matter-level permissions from document systems such as NetDocuments and iManage. Google names Harvey and Onit among the partners building on it.
How to access the Gemini agent
Access is limited for now. The companion post for smaller businesses states it directly. It says "Gemini is in early access with SMB customers today and will be available to all customers soon."
The product page gives two routes to try it:
- Contact Google Cloud sales to plan a company rollout.
- Start a 30-day trial. The product page says you can "try Gemini Enterprise Plus edition at no cost for 30 days without sales assistance."
Some features are further out. The product page says delegating multi-step background tasks, Gemini on mobile and desktop, and third-party model choice "are currently in early access."
On price, neither the announcement nor the product page lists a price for the Gemini agent. A Google Cloud pricing address for Gemini Enterprise returned a not-found page on October 9. Check with Google Cloud sales before you plan a budget.
Security and governance
Google built the agent's controls around four questions: who the agent is, what it may do, what it did, and what it should never touch.
- Identity. Each agent gets its own identity, governed like an employee's, with least-privilege permissions.
- Permissions. Security admins approve role-based access, and external connections pass identity through standards such as OAuth.
- Audit trail. Every action is logged and attributed to the agent rather than to a person.
- Sandbox and gateway. Agents run in an Agent Sandbox, and all their traffic passes through Agent Gateway.
On data use, the product page says customer data "is never used to train Google models or models for any other customer." That line covers prompts, outputs, and model tuning.
Alternatives worth comparing
Several vendors now sell an agent that takes a goal and returns finished work. The table compares them on what each vendor's own pages describe.
| Option | Where it works | What its pages describe | How to get it |
|---|---|---|---|
| Gemini agent | Web, mobile, desktop, Workspace, Microsoft 365, Slack | Objectives in, finished work out, with sub-agents, skills, and spend caps | Early access, or a 30-day Gemini Enterprise Plus trial |
| Claude Cowork | Desktop app, rolling out on web and mobile (beta) | Works across your files and tools, with scheduled tasks | Paid Claude plans |
| Copilot Cowork | Microsoft 365 | Long-running, multi-tool tasks that return a completed result | Microsoft 365 Copilot license plus usage |
| ChatGPT Work | ChatGPT on web, mobile, and desktop | Longer, multi-step work and finished deliverables | Eligible paid ChatGPT plans |
| Powerdrill Bloom | Its own workspace | Files and data in, sourced charts, reports, and slides out | Free and paid plans |
Claude Cowork. Anthropic's product page says "Give it a goal, and it works across your files and tools." It adds that you "come back to polished work for your review." Our comparison of Powerdrill Bloom vs. Claude Cowork goes deeper.
Copilot Cowork. Microsoft made it generally available on June 16, 2026. Its post says Cowork "executes complex, long-running, multi-tool tasks." It requires the Microsoft 365 Copilot license, and tasks are billed by usage in Copilot Credits. See Powerdrill Bloom vs. Copilot Cowork for details.
ChatGPT Work. OpenAI's help center describes Work as "an agent designed for longer, multi-step work and finished deliverables." Its examples include a document, spreadsheet, presentation, report, or Site.
Where a file-first workspace fits
The Gemini agent is built for companies whose data already lives in Google Cloud and Workspace. Some analysis starts somewhere smaller: a folder of exports, a PDF report, and a question due by Friday.
That is the job Powerdrill Bloom is built around. Its pricing page lists "Upload Excel, CSV, PDF, and docs" on the Free plan. Paid plans add "Create slides, Office docs, Excel analysis, and Nano Banana images." Its homepage calls it "The AI data analyst that shows its work."
Sourcing is the point. The homepage says "Every number comes back with the page, the row and the figure behind it." It also lists a Scheduled Agent, so "your recurring analysis runs itself." The AI report generator page covers turning that analysis into a written report.
How to evaluate the Gemini agent
Start with one task you already do every week, such as a project update deck or a weekly sales readout. You know what a good result looks like, so mistakes are easy to spot.
Then run three checks on the output:
- Source check. Trace three numbers in the output back to the system they came from.
- Format check. Confirm the result lands where your team reads it, such as a Doc, a deck, or an email.
- Control check. Set a spend cap on a test project and confirm the agent pauses when it hits the limit.
If the job starts from files rather than a connected warehouse, you can try Powerdrill Bloom on the same task and compare the two results.
Frequently asked questions
What is the Gemini agent?
The Gemini agent is Google's single agent for work, announced on October 8, 2026. It answers questions, handles knowledge work, creates images and media, and writes and runs code. You assign an objective, and it returns finished work in tools such as Docs, Sheets, and Slides.
How do I access the Gemini agent?
Google says the Gemini agent is in early access with small and midsize business customers and will reach all customers soon. You can contact Google Cloud sales, or start a 30-day Gemini Enterprise Plus trial without sales help.
How much does the Gemini agent cost?
Google's announcement and product page do not list a price for the Gemini agent. The product page directs buyers to Google Cloud sales or the 30-day trial. Built-in cost controls include per-project spend caps in the Cloud Billing Console.
Does the Gemini agent use Claude models?
Google's blog post says the agent orchestrates across Gemini models and Claude models from Anthropic today. The product page lists third-party model choice among features in early access, so availability may depend on your account.
Does the Gemini agent work in Google Docs and Sheets?
Yes. Google says it works inline in Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar, with the same skills and controls it has elsewhere. It can also act as a coworker agent that teammates mention in a Chat space or a document comment.
Sources: Google Cloud, Welcome to Gemini at Work 2026: Introducing the Gemini agent · Google Cloud, Empowering SMBs to do more with Gemini · Google Cloud, Gemini for Business · Anthropic, Claude Cowork · Microsoft 365 Blog, Copilot Cowork is now generally available · OpenAI Help Center, ChatGPT Work and Codex. Details read from vendor pages on October 9, 2026.