ReleaseBytes
gcp Google Cloud Blog ·

Build AI Agents on GCP with Gemini Enterprise Agent Platform Demos

blogaigcppreviewengineerretailgovernmentgcp-bigquerygcp-cloud-rungcp-pubsub
feature announcement

Google has released 13 hands-on demos for its Gemini Enterprise Agent Platform, showcasing how to build, scale, govern, and optimize AI agents. These demos utilize the code-first ADK and offer practical patterns for enterprise development, from basic agent creation to complex, long-running, and multi-agent pipelines. Developers can leverage these examples to integrate with services like BigQuery and Cloud Run, with features for security, governance, and optimization built-in.

  • →Agents CLI for Seamless Development Integration
  • →Event-Driven Approval Agent with Human-in-the-Loop
  • →Model Context Protocol for Data Connectivity
  • →Agent-to-UI (A2UI) for Dynamic Frontend Experiences
  • →Deploying Stateful Agents to Agent Runtime
Features (5) ›
  • Agents CLI for Seamless Development Integration

    The Agents CLI allows developers to integrate Agent Platform skills directly into their existing coding agents. This enables natural language descriptions for scaffolding, evaluation, deployment, and monitoring of agents without leaving the editor.

  • Event-Driven Approval Agent with Human-in-the-Loop

    A demo showcases an expense approval agent built with ADK 2.0's graph-based workflow API. It handles auto-approval for expenses below a threshold, applies security screens and compliance analysis for larger expenses, and includes a human review step before finalization, triggered via Pub/Sub events.

  • Model Context Protocol for Data Connectivity

    The Model Context Protocol (MCP) codelab demonstrates building reusable tools that enable Gemini to query BigQuery, search files, and call APIs. MCP is an open protocol designed for cross-vendor and cross-framework compatibility.

  • Agent-to-UI (A2UI) for Dynamic Frontend Experiences

    This demo focuses on building agents that can dynamically render and update user interface components in real time as a conversation progresses. The agent assembles the necessary UI elements on the fly, creating highly visual user experiences.

  • Deploying Stateful Agents to Agent Runtime

    Demos cover deploying stateful data science agents to Agent Runtime (formerly Agent Engine), enabling automatic infrastructure management, scaling, and session management via features like Memory Bank. Additionally, agents are shown to handle long-running enterprise workflows reliably by pausing, resuming, and maintaining context across restarts using durable state machines and event-driven patterns.

Notes (1) ›
  • Introduction to Gemini Enterprise Agent Platform and Demos

    Google introduced the Gemini Enterprise Agent Platform earlier this year and has now released 13 demos to illustrate its capabilities. These demos provide practical concepts, patterns, and architectures for building AI agents.

Read the original announcement →

https://cloud.google.com/blog/products/ai-machine-learning/13-demos-on-gemini-enterprise-agent-platform/

Related releases