Google launches Gemini Enterprise for IT teams, targeting agent governance

Google is making a clear wager on who will control AI inside the enterprise: not every employee, but the IT teams that already decide what gets deployed, monitored and trusted. At Cloud Next 2026 on April 22, the company introduced Gemini Enterprise Agent Platform as a system to build, scale, govern and optimize agents, presenting it as the next stage of Vertex AI and as infrastructure for organizations trying to manage AI at scale.

The platform is aimed first at technical teams and developers, not general office workers. Google says it adds agent integration, DevOps, orchestration and security features, positioning the product as a developer platform and a core control layer for the agentic enterprise. That framing matters because enterprise agents are still new, security concerns remain high, and many companies are only beginning to figure out how to approve, deploy and audit them.

Google’s pitch also makes the governance angle explicit. The company says Gemini Enterprise is deeply integrated with its data and security tools and is built for organizations that need to govern and optimize thousands of agents, not just test a handful of pilots. Admins can control which partner-built agents are available through an Agent Gallery, while a separate Agent Marketplace is part of the broader ecosystem. The Gemini Enterprise app sits alongside that infrastructure layer for employees who need to discover, create, share and run agents in one secure platform.

The app is designed to connect with the systems businesses already use, including Google Workspace, Microsoft 365, Salesforce and SAP. Google says the platform includes enterprise-grade security and compliance, a crucial selling point for companies wary of letting AI touch internal files, workflows and customer data. The company also highlighted access to Gemini 3.1 Pro, Gemini 3.1 Flash Image, also called Nano Banana 2, and Lyria 3, underscoring that it is offering multiple model types rather than forcing customers into a single stack.

That multi-model approach reflects the competitive reality Google is entering. The company is taking on Amazon’s Bedrock AgentCore and Microsoft Foundry, while also allowing Anthropic’s Claude family, including Opus, Sonnet and Haiku, to sit underneath parts of the platform. In March 2026, Google said scaling agents beyond company walls requires standardized protocols, shared trust and agent-ready governance. In April, it said the conversation had shifted from whether an agent could be built to how thousands of agents could be managed.

Google Cloud is backing that argument with infrastructure scale. It says its Cross-Cloud Network is used by 65% of the Fortune 100 and handles up to 27 exabytes of data per month. The company also said Agentspace was folded into Gemini Enterprise, with its orchestration technology now powering core functions. The message is straightforward: enterprise AI will not spread mainly through bottom-up experimentation. Google wants it to move through IT, as governed plumbing that large organizations can standardize, secure and control.