TLDR: Google’s Firebase Studio has introduced new ‘Agent Modes’ powered by Gemini 2.5, designed to significantly automate and accelerate coding tasks. Unveiled at the Google Cloud Summit London, these modes offer developers three levels of AI collaboration: conversational ‘Ask’ mode, human-in-the-loop ‘Agent’ mode requiring approval for changes, and an autonomous ‘Agent (Auto-run)’ mode capable of independently writing code, fixing errors, and building features, aiming to reduce manual development effort and enhance productivity.
At the recent Google Cloud Summit in London on July 10, 2025, Google announced significant advancements to Firebase Studio, its cloud-based AI workspace for creating full-stack AI applications. The highlight of these updates is the introduction of versatile ‘Agent Modes,’ alongside foundational support for the Model Context Protocol (MCP) and Gemini CLI integration, all aimed at redefining AI-assisted development and embedding powerful AI capabilities directly into the developer workflow.
Firebase Studio, a unification of Google Project IDX and Gemini, serves as an integrated and extensible agentic workspace for building, running, and managing web apps, cross-platform mobile apps, and backend services. It allows developers to import existing projects or start from scratch using templates that support various programming languages and frameworks. The core innovation lies in its new autonomous Agent Mode, which leverages the deep code understanding and powerful reasoning capabilities of Gemini 2.5.
Developers now have three distinct levels of AI collaboration within Firebase Studio:
1. Ask Mode: This conversational mode is designed for discussion and planning with Gemini. It’s ideal for brainstorming, planning code, and collaboratively discussing complex problems without making any changes to project files.
2. Agent Mode (Human-in-the-Loop): In this mode, Gemini can propose changes to an application, but developers retain complete oversight. Any proposed modifications must be explicitly approved by the developer before files are altered, allowing for thorough code review and control over integration into the project.
3. Agent (Auto-run) Mode: This is the most powerful and autonomous mode, designed to significantly accelerate development. Gemini can autonomously reason, apply code changes across multiple files, write tests, fix errors, and build out features or refactor components. For security and control, this mode still requires explicit permission before performing sensitive actions such as deleting files, running terminal commands, or using external tools.
These Agent Modes can be seamlessly toggled, allowing developers to choose the level of AI assistance best suited for their current task. They also leverage personalized guidance from project-level rule files, such as .idx/airules.md, GEMINI.md, or .cursorrules, ensuring Gemini adheres to specific design patterns and preferences. This automatic detection and loading of instructions provide a cohesive and highly customizable experience.
Also Read:
- Google DeepMind Unveils Gemini Robotics On-Device: Empowering Robots with Local AI Capabilities
- Google Unveils Vertex AI ‘Memory Bank’ for Persistent AI Agent Context
Google Cloud emphasizes that these advancements are part of a broader strategy to transform development with AI, enabling developers to build the next generation of agentic applications. While the current capabilities allow for the creation of relatively simple applications, the pace of development is impressive, and the complexity of what can be created is steadily increasing. The integration of Gemini CLI further streamlines the development process, allowing developers to interact with Gemini directly from their command line. Additionally, the foundational support for the Model Context Protocol (MCP) will enable developers to extend and personalize their workflows with Gemini within Firebase Studio by adding MCP servers to their workspace.
