谷歌推出Gemini Enterprise Agent Platform，Vertex AI全线并入

ME News, April 22 (UTC+8). According to Beating monitoring, Google launched the Gemini Enterprise Agent Platform at Cloud Next 2026, integrating the previous Vertex AI in full, and all subsequent Vertex AI services and roadmaps will be delivered through this platform. The platform covers the end-to-end process of agent building, scaling, governance, and optimization, positioning itself as the control center for enterprise agent clusters.

The platform includes several new components. The Agent Development Kit (ADK) adds a graph framework that allows orchestrating multiple agents into a sub-agent network. Agent Runtime has been rebuilt for production environments, supporting long-running agents to maintain state for days, and introducing Memory Bank to provide persistent long-term context. In the governance layer, three modules—Agent Identity, Agent Registry, and Agent Gateway—are added to respectively manage agent permissions, registration tracking, and security guardrails. Agent Sandbox provides an isolated environment for securely executing model-generated code and browser automation tasks.

At the model layer, the platform supports Google’s own models (Gemini 3.1 Pro, Gemini 3.1 Flash Image, Lyria 3), the open-source model Gemma 4, and third-party models such as Anthropic Claude. Pichai also announced a set of business metrics. The paid monthly active users of Gemini Enterprise grew 40% quarter-over-quarter in the first quarter. Customer direct calls via API to Google’s one-side models reached 16 billion tokens per minute, up from 10 billion in the previous quarter. In 2026, more than half of the company’s machine learning compute investment will be directed to cloud business.

On the security front, Google released a set of agent-driven threat detection solutions that integrate its Threat Intelligence and Security Operations with Wiz’s cloud security platform. Wiz also simultaneously rolled out the AI Application Protection Platform (AI-APP), covering autonomous protection from code to cloud to runtime, supporting multi-cloud, hybrid, and AI environments.

(Source: BlockBeats)