Google Launches Gemini Enterprise Agent Platform for Long Running Autonomous Agents

Google DeepMindGoogle DeepMind

· Updated

Google launched the Gemini Enterprise Agent Platform, a unified environment that replaces Vertex AI for building, scaling, and governing AI agents. The platform introduces infrastructure for agents that can run for multiple days and maintain persistent memory, moving AI from reactive chat to autonomous digital employees.

Google launched the Gemini Enterprise Agent Platform, a comprehensive system for developing and managing autonomous agents. It evolves Vertex AI by adding an Agent Development Kit (ADK) for building multi-agent networks and an Agent Runtime designed for sub-second starts and multi-day task execution.
Model access
200+ models including Gemini 3.1 and Claude
Development tools
Agent Studio and Agent Development Kit
Persistence
Agent Memory Bank for long-term context
Execution
Agent Runtime supporting multi-day workflows
Security
Agent Identity and isolated Agent Sandbox
Integration
Native Model Context Protocol support

This shift addresses reliability gaps in agentic AI by providing a Memory Bank for persistent context and Agent Identity for cryptographic auditing. Organizations can now deploy agents that handle complex, multi-step workflows like sales prospecting while maintaining enterprise-grade security and governance guardrails.

You can build agents using the low-code Agent Studio or the code-first ADK, with access to over 200 models including Gemini 3.1 Pro. The platform is generally available through the Google Cloud console, featuring an Agent Sandbox for safe code execution and support for the Model Context Protocol.

View the full update on cloud.google.com
Google DeepMind
Google DeepMind
@GoogleDeepMind
X

We’re launching Gemini Enterprise Agent Platform with @GoogleCloud: a platform for businesses to develop, scale, govern and optimize agents. It’s the evolution of Vertex AI, bringing together model selection and agent building with new features for integration, security and more.

146retweets1.3klikes
View on X

Still wondering? A few quick answers below.

The Gemini Enterprise Agent Platform is a unified environment from Google Cloud designed to build, scale, govern, and optimize AI agents. It serves as the evolution of Vertex AI, providing technical teams with tools to create autonomous digital workers that can execute complex, multi-step business processes with enterprise-grade security and reliability.

While Vertex AI focused on model development and basic AI tasks, the Gemini Enterprise Agent Platform is specifically re-engineered for the agentic era. It introduces new infrastructure like the Agent Runtime for long-running tasks and the Agent Development Kit for multi-agent orchestration, moving beyond simple chat interactions to autonomous execution.

Agent Memory Bank is a persistent storage system that allows AI agents to generate and curate long-term memories from past conversations. This enables agents to recall high-accuracy details and user-specific context with low latency across different sessions, ensuring that the AI maintains a deep understanding of user history and preferences over time.

Yes, the platform includes an Agent Sandbox that provides a hardened, isolated environment for agents to safely execute model-generated code and perform computer use tasks like browser-based automation. This allows agents to take real-world actions and manage background operations without posing a security risk to the host enterprise systems.

The platform provides access to over 200 leading models through the Model Garden. This includes Google's first-party breakthroughs like Gemini 3.1 Pro, Gemini 3.1 Flash Image, and Lyria 3, as well as open models like Gemma 4. It also supports third-party models from providers like Anthropic, including Claude Opus, Sonnet, and Haiku.

Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →

Google

See all AI news & updates from Google →

Share this update