Google Cloud's Gemini agent is a single enterprise AI agent that works like a colleague
Google Cloud unveiled the Gemini agent at Gemini at Work 2026: one universal agent for knowledge work, coding, and data that can hold its own Workspace account.
At its Gemini at Work 2026 event on 8 October 2026, Google Cloud CEO Thomas Kurian introduced the Gemini agent: one universal AI agent for enterprise work. Not a suite of specialised tools you stitch together, and not a chatbot you interrogate for answers. The idea is simpler and more ambitious than either: you describe what you need done, and the agent goes away and does it.
“You give it objectives, not instructions,” Kurian said during the keynote. “You delegate an outcome and come back to finished work.”
What the Gemini agent actually does
The Gemini agent handles knowledge work, question answering, content creation, and coding from a single prompt interface. It works inline inside Gmail, Drive, Docs, Slides, Sheets, Chat, and Calendar, and it carries the same memory, skills, and controls regardless of where you interact with it. You can also reach it through the command line, Microsoft 365, Slack, or third-party applications, no dedicated interface required.
What separates this from a capable assistant is persistence. The agent runs in the cloud, so tasks that take hours or days continue even after you close your laptop. For larger jobs, it can spin up temporary sub-agents, each with its own identity and dedicated expertise, then coordinate their outputs back into a finished result. Maximum runtime for a single task sits at seven days.
The virtual colleague feature is worth pausing on
One of the more striking capabilities announced is the ability to create a coworker agent that functions within your team like a staff member. You describe the role you need, and Gemini creates an agent that receives its own Workspace account: an email address at your company domain, a calendar, Drive storage, and an entry in your company directory. Colleagues can add it to a Chat space or @mention it, the same way they would anyone else.
Each agent identity is cryptographically attested and governed with least-privilege permissions. When the agent connects to an external system, that identity is propagated through OAuth. Every action it takes is stamped into an audit log. IT administrators can visualise and audit all agent activity across the organisation.
For teams that have been manually orchestrating multiple AI tools across different interfaces, this is a meaningful consolidation. There is one place to govern permissions, one identity framework to manage, and one audit trail to review.
New data, analytics, and industry skills
The Gemini agent ships with new data and analytics capabilities built on BigQuery and Knowledge Catalog. Business users can ask questions in plain language, and the agent identifies the relevant data sets, writes the necessary SQL, Spark, or Python, and runs it. Technical teams get the same starting point but with more headroom to direct the process.
Two industry specialisations are now in preview. Gemini for Financial Services connects to FactSet, LSEG, Moody’s, MSCI, PitchBook, S&P Global, and the SEC’s Edgar system, among 13 financial data connectors in total. It covers investment research, credit analysis, and risk modelling, and surfaces confidence scores, methodology notes, and full data lineage to satisfy audit requirements. BNP Paribas is already deploying it across more than 65,000 employees in its Corporate and Institutional Banking division.
Gemini Enterprise for Legal is built around the specific requirements of legal teams: matter-level permissions and ethical walls inherited directly from document management platforms like NetDocuments and iManage. It handles contract review, diligence, regulatory monitoring, and privacy requests, with partner agents and connectors purpose-built for legal workflows.
Government, healthcare, and retail specialisations are listed as coming soon.
What this means for you
If you work in a knowledge-heavy role, the practical shift here is moving from “ask AI for help” to “assign AI a task”. The Gemini agent is designed to sit in your team’s workflow the way a contractor or junior colleague might: you hand it a brief, it has access to the systems it needs, and you review the output rather than doing the work yourself.
If you manage IT or security for an organisation, the governance story matters as much as the capability story. The Agent Gateway enforces organisation-wide policies on communications between agents and external systems. Sandboxing puts boundaries around code execution. Role-based access permissions are approved by your security administrators before any agent connects to external systems. That is a more structured security model than most organisations have had for AI tools to date.
If you are evaluating this for financial services or legal, the compliance-first design is deliberate. Both specialisations are built to show their working: sources, methodology, data lineage, and confidence scores are part of the output, not an afterthought.
Connectors and model flexibility
The Gemini agent connects to Salesforce, ServiceNow, Jira, Git, BigQuery, Snowflake, Databricks, and Microsoft Teams, among others. It also supports Model Context Protocol (MCP) for connecting to additional external tools and data sources.
On the model side, Google has built in routing so that simple tasks run on Gemini Flash and more complex ones escalate to more capable models. The system also supports multi-model routing including Claude from Anthropic, which is a notable choice given the competitive context.
Availability
The Gemini agent is in private preview as of 8 October 2026, with wide availability coming soon for Workspace customers on select Business and Enterprise plans. Google Cloud noted that nearly 90% of the Fortune 100 are already using Gemini Enterprise, and 500 customers each processed more than one trillion tokens over the past year, which gives some sense of the deployment scale this is being built on top of.