Google Cloud is launching the Gemini Enterprise Agent Platform, an evolution of its Vertex AI, to serve as a central hub for building artificial intelligence agents. Announced at Google Cloud Next 2026, the platform integrates model selection, development tools, and new features for agent integration, orchestration, DevOps, and security.
The new offering aims to provide a single destination for technical teams to develop AI agents that can be deployed to employees via the new Gemini Enterprise application, enabling widespread work automation.
Michael Gerstenhaber, Google Cloud VP of Product Management, stated the platform is designed to handle the complexity of agents interacting across multiple systems securely and with governance. The Gemini Enterprise Agent Platform covers the entire lifecycle of AI agents, broken down into four pillars: Building, scaling, governing, and optimizing autonomous workforces.
Key features for building include Agent Studio, a low-code interface for business users, and the Agent Development Kit (ADK) for developers. The ADK allows for more powerful reasoning, organizing agents into networks, and native ecosystem integrations. For scaling, the platform offers a revamped Agent Runtime for simple provisioning, support for multiday workflows, and tools for agent-to-agent orchestration. A new Agent Memory Bank dynamically generates long-term memories from conversations for recall.
Governance is handled through a secure-by-design architecture with enterprise policy controls, unique cryptographic Agent Identities for auditable actions, and an Agent Registry for a central library of approved tools. The Agent Gateway acts as a control tower for overseeing AI agent fleets and enforcing security policies. Tools for protecting against prompt attacks and real-time monitoring are available via the Agent Security dashboard.
For optimization, the platform includes Agent Simulation for pre-shipment testing and Agent Evaluation for scoring performance in production. Agent Observability allows tracing complex reasoning, and the Agent Optimizer can refine system instructions to enhance accuracy.
While Google will promote its Gemini models, the platform supports an open model ecosystem, offering access to over 200 models including Gemini, Gemma, Lyria, and third-party options like Anthropic's Claude.