Google Cloud has launched a universal AI agent that doesn't just assist employees — it joins the team. Announced at the company's Gemini at Work 2026 event, the new Gemini agent can receive its own Workspace account, email address, calendar, Drive storage, and entry in the company directory. It plans work, delegates to sub-agents, connects to more than a dozen enterprise systems, and keeps running for days after the laptop closes. Australian retail conglomerate Wesfarmers is among the early testers.
From assistant to colleague
The shift in language is deliberate. Google Cloud CEO Thomas Kurian framed the agent as something that receives "objectives, not just instructions" — it plans the work, selects the tools, and brings back finished output. A manager could create a Gemini "Event Planner Agent" that receives its own @agents.company.com email, appears in the company directory, and can be tagged in Google Chat, added to group conversations, or emailed directly — just like a human colleague.
This is a meaningful architectural choice. Every action the agent takes is written to an audit trail attributed to the agent, not the person who created it. The identity is cryptographically attested using the SPIFFE standard, with role-based permissions, OAuth delegation for external systems, and what Google calls an "AI network firewall" that enforces organisation-wide policies on agent-to-agent and agent-to-external communications.
The practical consequence: if the agent sends an email, books a meeting, or modifies a spreadsheet, the company knows exactly which agent did it, under what authority, and who is accountable.
What the agent actually connects to
The integration list is broad. Beyond Google Workspace, the agent works with Microsoft 365, Slack, Jira, Confluence, Git, BigQuery, Databricks, Snowflake, Salesforce, ServiceNow, and Postgres. It supports Model Context Protocol (MCP) servers and internal tool registries.
Perhaps most notably, it isn't locked to Google's own models. The agent can route tasks to Anthropic's Claude alongside Google's Gemini models, with support for additional proprietary and open-weight models planned. A Smart Routing capability selects models based on performance and cost per task — expensive models for hard problems, cheaper ones for routine work.
For cost-conscious businesses, Google is also introducing project-level spending limits that can pause agent activity when budgets are exhausted, with cost attribution by department. After stories like Uber's AI budget blowout earlier this year, these controls matter.
The numbers behind the push
Google CEO Sundar Pichai opened the event with some striking adoption figures. Gemini now has over 1 billion monthly active users. Nearly 90% of Fortune 100 companies use Gemini Enterprise at work. And nearly 80% of all Google Cloud customers are actively using AI products, with almost 500 enterprise customers each processing over one trillion tokens in the past year.
Those are platform-wide numbers, not specific to the new agent. But they explain why Google is betting enterprise customers will trust a general-purpose agent with cross-system access: the existing user base is already deeply embedded.
Among early testers, the results are concrete. Sportswear company On cut enterprise service migration time from three months to two weeks per service after adopting an agent-led strategy. Japanese insurer SOMPO has built more than 10,000 custom agents. BNP Paribas is deploying Gemini Enterprise to over 65,000 employees.
The Australian angle: Wesfarmers goes all-in
For Australian businesses watching this space, the Wesfarmers story is worth paying attention to. The company — which owns Kmart, Officeworks, Priceline, and Bunnings — signed a multi-year collaboration with Google Cloud in February to deploy Gemini Enterprise across its retail and health businesses. It's already piloting cross-brand conversational shopping through OnePass, where customers can search and buy across multiple Wesfarmers brands in a single conversation.
Wesfarmers Managing Director Rob Scott described the scope: "We are continuing to invest in AI across our businesses, with expanding use in areas including conversational and agentic commerce, team member AI assistants, merchandising and marketing effectiveness, customer self-service and supply chain optimisation."
Notably, Wesfarmers isn't betting on a single provider. It also signed a multi-year partnership with Microsoft to scale Azure OpenAI and Microsoft 365 Copilot across Bunnings, Kmart Group, Blackwoods, and Priceline. That multi-vendor approach — Google for Workspace-native agents, Microsoft for the 365 estate — increasingly looks like the default enterprise playbook for 2026.
The competitive context: everyone wants to be the central agent
Google's announcement didn't happen in a vacuum. In the two weeks before, Microsoft introduced an updated Copilot combining productivity, coding, and persistent automation through its Autopilot agent. OpenAI launched its Dots agents at DevDay. Anthropic announced native Claude integration with Google Docs, Sheets, and Slides. SpaceXAI expanded its Grok Bot with shared Team Bots.
The pattern is clear: every major AI company now wants to provide the central operating system for knowledge work, not just a chatbot you query when you're stuck. As CCS Insight noted, the real test is "whether employees can entrust it with useful work and spend less time managing that work themselves."
For businesses already on Google Workspace, the Gemini agent has a structural advantage — it sits inside the tools people already use. For Microsoft 365 shops, Copilot holds the same position. For teams that need heavy coding or long-context analysis, Claude continues to win that lane. Most enterprises are likely to end up running two or three of these systems, not one.
What to watch
The agent is currently in private preview with general availability expected late October or early November. Google hasn't announced standalone pricing — it's unclear whether the agent will be included in existing Gemini Enterprise subscriptions or require an add-on.
The industry-specific versions are worth tracking. Financial Services includes more than 50 foundational skills and connections to FactSet and LSEG. The Legal version focuses on document-management permissions and confidentiality with support for NetDocuments and iManage. If Google can deliver agents that understand the compliance requirements of specific industries — not just general-purpose task completion — that's where the real enterprise value lies.
For smaller businesses, the immediate question is simpler: do you want an AI that works like a team member, or one that works like a tool? Google is betting on the former. The governance architecture — identity, audit trail, permissions, spend caps — suggests they've thought carefully about what happens when autonomous software acts inside a business without someone watching every step.
At Heygentic, we think the identity layer is the most significant part of this announcement. The capabilities are impressive, but every vendor has capable models. What Google is really building is the accountability infrastructure that enterprises need before they'll trust an agent to send an email, file a report, or book a meeting on their behalf. That's the hard problem — and the one that will determine whether AI agents become real workers or remain clever demos.
Sources
- Google Cloud launches Gemini agent — Google Blog
- Google brings agentic AI to Gemini, starting with businesses — TechCrunch
- Google Cloud unveils persistent Gemini Agents for long-running tasks — VentureBeat
- Google launches workplace AI agent that gets its own email address — The Next Web
- Google's Gemini Agent Puts Workplace Delegation to the Test — CCS Insight
- Google Cloud launches Gemini agent to work across enterprise systems — Constellation Research
- Wesfarmers strengthens AI capabilities with Google Cloud Collaboration — Wesfarmers
- Wesfarmers to deploy agentic AI groupwide — iTnews
