Cloud Stage - Day 1 (Google I/O 2025)

Key Concepts

• Vertex AI: Google Cloud's AI platform for building, training, and deploying models and agents.
• Gemini: Google's family of multimodal AI models.
• Agent Development Kit (ADK): An open-source framework for building AI agents in Python and Java.
• Model Context Protocol (MCP): A protocol for agents to interact with tools and data sources.
• Agent2Agent Protocol (A2A): A protocol for agents to communicate with each other.
• Agent Engine: A managed service for running and scaling AI agents.
• Agentspace: An enterprise platform for deploying and managing AI agents for internal use.
• Gemini Code Assist: A coding assistance tool that provides code completion, generation, and training.
• Firebase Studio: A cloud-based AI workspace for building and deploying applications.
• Firebase AI Logic: A tool for integrating, innovating, and deploying AI seamlessly in apps.
• Genkit: An open-source toolkit for building powerful backends with AI.
• Firebase Data Connect: Connects your application to the cloud scwt through security so everyless APIs.
• Firebase App Hosting: A solution for hosting modern web applications with server-side rendering.
• Cloud Run: A serverless platform for running containerized applications.
• BigQuery: Google's fully-managed, serverless data warehouse.
• GKE: Google Kubernetes Engine.
• TPU: Tensor Processing Unit.

AI Models and Vertex AI

• Model Updates: Gemini 2.5 Pro and Flash are coming soon in preview. VO3 (video with audio creation) and Imagen 4 (image generation) are available on Vertex. Veo 2 and Llama 4 are generally available. Llama 4 is offered as a service, with users paying only for tokens consumed.
• Vertex AI Optimizer: Automatically selects the most cost-effective or fastest model for a given request.
• Pretraining Options: Supports supervised fine-tuning (using labeled data to improve model context), Retrieval-Augmented Generation (RAG) using external data, full fine-tuning, and in-context learning (leveraging long context windows).
• Vertex AI Global Endpoint: Simplifies model deployment by allowing users to specify a global endpoint URL instead of region-specific ones, optimizing for availability and performance (with caveats for context caching).

Agent Building and Deployment

• Vertex AI Agent Builder: A suite of tools for building, deploying, and managing AI agents, including the Agent Development Kit (ADK), Agent Garden, and integration with other tools.
• Agent Development Kit (ADK):
  • Version 1.0 released for Python, with a new Java version available.
  • Framework for building simple or complex multi-agent applications.
  • Supports no-code agent creation using instructions, models, and tools.
  • Model-agnostic, allowing integration of non-Gemini models.
  • Supports MCP for tool integration and A2A for agent communication.
• Agent2Agent Protocol (A2A): Enables agents to communicate and share capabilities. An SDK is available in Python. Microsoft has adopted this protocol.
• Agent Engine: A managed service for running, scaling, and observing AI agents. Includes a new visual dashboard for easier management. Supports deployment from the ADK via CLI.
• Agentspace: An enterprise platform for deploying and managing AI agents for internal use, connecting to employee directories, third-party systems, and Google Workspace services.

AI at the Edge

• Gemma 3: A small, high-performing model for mobile devices and laptops, now available.
• On-Device RAG: Enables retrieval-augmented generation on mobile devices.

App Development

• Gemini Code Assist:
  • Free for individual use.
  • Provides code completion, generation, and training on custom codebases.
  • Integrates with Google Docs, GitHub, and other tools via the @ symbol.
  • Supports custom rules for code generation (e.g., always generate unit tests).
  • Maintains a history of chats and prompts.
• Firebase Studio:
  • Cloud-based AI workspace for building and deploying applications.
  • Supports Go and Python backends, with over 60 templates for common languages and frameworks.
  • Integrates with Figma via builder.io.
  • Uses Gemini 2.5 for more complex app creation.
  • Offers backend integrations with Firestore and Firebase Auth.
• Application Design Center: A tool for visually modeling and deploying cloud systems.
• Cloud Hub: Provides an app-centric view of application health, cost, and deployment information.
• Gemini Code Assist Integration: Provides AI-powered troubleshooting and optimization within Google Cloud.
• Cloud Run:
  • One-click deployment from Google AI Studio.
  • No-code Cloud Run MCP server for integration with various tools.
  • Supports Gemma 3 deployment with GPUs and scale-to-zero capabilities.

Security

• New unified security offerings and integrations in Chrome.
• DNS armor for DNS protection.

Data and Analytics

• MCP Toolbox for Databases: An open-source project for connecting to various databases (including Google Cloud and third-party) with a common contract.
• Firestore MongoDB Compatibility: Offers a MongoDB instance on Firestore with serverless scaling.
• Database Center: Provides a global view of database health and management across Google Cloud databases.
• BigQuery:
  • AI-powered code completion, explanation, and optimization.
  • Vector search capabilities.
  • AI-assisted notebooks.
  • Business glossary for common terms.
  • Metadata export to Cloud Storage.
  • At-scale categorization of objects in Cloud Storage.
  • Disaster recovery capabilities.
  • Apache Iceberg support.
  • Serverless Spark integration.
  • Real-time query processing on streaming data.
  • History-based query optimization.
  • Pipe syntax support.
  • Geospatial analytics.
  • Conversational analytics using natural language.
  • Spectacles for Looker dashboard delivery.

Infrastructure

• Cloud Wide Area Network (WAN): Allows leveraging Google's network backbone for WAN connectivity.
• Cloud Run GPUs: Provides on-demand GPUs for serverless AI workloads, generally available with no quota request.
• GKE Inference Gateway: Enables intelligent routing, model selection, and observability for models served on GKE.
• Gemini on Google Distributed Cloud: Makes Gemini available on-premises, either connected or air-gapped.
• Distributed Inference with VLLM: A project with Red Hat, IBM, and CoreWeave to support distributed inference using VLLM on Kubernetes.
• Ironwood: A new 7th-generation TPU with improved performance.

Firebase

• Firebase Studio: Cloud-based AI workspace for building and deploying applications.
  • 1. 5 million workspaces created since launch.
  • Integrates Gemini 2.5 for complex app creation.
  • Backend integrations with Firestore and Firebase Auth.
  • Partnership with builder.io for Figma design integration.
• Firebase AI Logic: Easiest ways to integrate, innovate, and deploy AI seamlessly in the apps.
  • Experimental support for hybrid inference in Chrome browsers (local Gemini Nano or cloud).
  • Unity support, including XR.
  • Expanded AI monitoring (latency, error rates, tokens used, traces).
• Genkit: Open-source toolkit for building powerful backends with AI.
  • Python and Go support.
  • Dynamic Model Registration (no need to wait for updates to use latest models).
  • Genkit.dev: New site for documentation and code examples.
• Firebase Data Connect: Connects your application to the cloud scwt through security so everyless APIs.
  • Transactions with additional server values and multisafe capabilities.
  • Schema and query generation agent (using Gemini).
• Firebase Phone Number Verification: Moves end users from an SMS-based workflow, which is somewhat errorful, to a method where with their permission, you can surrender fight phone number directly with the carrier.
• AI Testing Agent: Systematically tests applications by simulating adversarial tasks and measures and benchmarks your LLM performance while it's doing it.
• App Testing Agent: Uses Gemini to create tests for mobile applications using natural language.
• Firebase App Hosting: Solution for hosting modern web applications with server-side rendering.
  • Deploying from a local Firebase CLI.
  • Terraform support.
  • Automation configuration of the Firebase SDKs them later suite.
  • Seamlessly integrates with Firebase Studio.
• Firebase MCP support: Now, you can use the Firebase MCP server from the CLI and use it to access the power of Firebase services from your favorite agents and soon from Firebase Studio too.

Conclusion

Google Cloud is heavily investing in AI, providing a comprehensive suite of tools and services for building, deploying, and managing AI-powered applications. The focus is on making AI more accessible and easier to use, with features like the Vertex AI Optimizer, Agent Development Kit, Gemini Code Assist, and Firebase Studio. The platform also emphasizes security, data management, and infrastructure optimization, offering a complete solution for developers and enterprises looking to leverage the power of AI.