Particle.news
Download on the App Store

Google Turns Agents Into the Core of App Development with Gemini Spark and Antigravity 2.0

The company has built an agent-native platform that stitches faster cloud models, managed persistent agents, orchestration tools, and on-device open models into a single development and runtime stack.

Overview

  • Google used its Google I/O keynote on Tuesday, May 19 to announce a coordinated agent stack that includes Gemini 3.5 Flash, Gemini Omni, Gemini Spark, Antigravity 2.0, Google AI Studio, Firebase agent features, and the Gemma 4 on-device models.
  • Gemini Spark is a managed, always-on personal agent that runs on Google Cloud and can act on users’ Gmail, Docs, Calendar, and other services so tasks run in the background even when devices are off.
  • Antigravity 2.0 and Google AI Studio compress the developer lifecycle by letting teams spin up parallel subagents, prototype in the browser, preserve context across handoffs, and export directly to production toolchains.
  • Firebase received agent-native primitives—Agent Skills, Firestore Agent‑Sync for persistent sessions and automatic context compression, and App Check intent protections—that make stateful, action-capable agents easier to build and secure.
  • Google also published on-device options with Gemma 4 and the AI Edge Gallery that run privately on phones using the Model Context Protocol, and early coverage says developers are seeing big productivity gains alongside new tradeoffs over control, privacy, and vendor lock-in.