Google opens Android to AI agents and strengthens Gemini

Google opens Android Studio to third-party AI agents, adds new Gemini APIs, and prepares connected app features. The whole aims at developers but also public uses in Search, Gemini, and Chrome.

On October 28, 2026, at the Android Dev Summit, Google unveiled a series of tools that place AI agents at the heart of Android, Gemini, and Chrome. The group is opening Android Studio to external agents, launching new APIs to execute functions on mobile, and preparing a more direct bridge between apps, Search, and the Gemini app. The goal is simple: to accelerate development and also make Google services more connected to real-world usage.

Android Studio devient une passerelle pour des agents tiers

With Android Studio Rabbit 2, Google introduces the "Bring Your Own Agent" feature, which allows the use of more than 40 third-party coding agents via the open ACP protocol, including Claude, Codex, and Google Antigravity. These agents have direct access to IDE tools such as the project structure, emulator, or Logcat, changing the way code is debugged and written. The promise is to reduce time spent on repetitive tasks and also the token cost.

The manufacturer also adds Change Review, a change verification tool that analyzes pending changes, generates bug reports, and explains differences step by step. In the same vein, the Canary version of Android Studio is testing a migration assistant capable of automatically converting old XML layouts to Jetpack Compose. Google is also experimenting with Lightbuild, a declarative build system based on a YAML schema, readable by both humans and agents.

Des outils pensés pour le terminal, pas seulement pour l’IDE

Google is also expanding Android CLI, its command-line tool, with official plugins for Claude Code, OpenAI Codex, and Google Antigravity. Agents can now generate Compose previews directly in the terminal without launching the emulator, and Google states that its internal tests show UI tasks completed 77% faster, with up to three times fewer tokens. Another feature allows streaming physical devices from Google labs to track hardware-dependent bugs.

The group also integrates LeakCanary, its memory leak detection library, into Android CLI and the Android Studio Profiler. According to Google, an A/B test conducted with Claude Opus 5.5 found and fixed memory leaks nearly twice as fast while reducing token usage by about 25%. Agents also have access to more than 30 "Android Skills," modules providing up-to-date documentation on Google Play policies, permission security, or interfaces adapted to foldables and Googlebooks laptops. Yes, machines now read the docs; developers can breathe a little.

Gemini 4 Carbon, Connected Apps et Chrome gagnent en portée

Meanwhile, Google announces Gemini 4 Carbon, presented as a next-generation model with improvements in code, agent tasks, cybersecurity, and multimodal understanding. According to Google, Carbon ranks first on version 5.0 of the Artificial Analysis Intelligence Index with a score of 59, moves up to 4th place in WebDev Arena with 1769 ELO points, and shows better results on SEC-bench Pro, ExploitGym, and CWE-bench. The model will first be offered to cybersecurity defenders this week, before wider availability for developers, businesses, and consumers [to be confirmed on the exact schedule].

On the applications front, Google launches the Connected Apps Platform, a unified gateway to connect more services into Gemini and Search. More than 50 integrations are already active in the Gemini app, with Adobe, Dropbox, Instacart, OpenTable, Ticketmaster, and Calm coming soon, and the platform will allow publishing an integration to Search, Gemini, or both. Developers will need to use an MCP server hosted in the cloud or Android MCP via the AppFunctions API, then manage everything from the Google Connected Console.

Chrome also receives a more discreet but very concrete new feature: "Suggested by Gemini" autofill. In the United States, Gemini users in Chrome can see suggestions drawn from Gmail to fill out forms, with encrypted sensitive data and full control in settings. Google thus extends the logic of its assistants to everyday life, from the terminal to the browser, through the mobile app and glasses.

Des lunettes XREAL AURA aux API d’apps : Google pousse l’IA partout

The hardware part of the package is called XREAL AURA, connected glasses running Android XR developed with XREAL. The model, sold for 1,099 dollars in the United States, Canada, and South Korea, weighs less than 95 grams, displays up to five virtual windows in the field of view, and relies on an external box that also functions as a trackpad. Gemini sees both the real environment and virtual screens to organize the workspace or locate physical objects.

Finally, Google presents new building blocks to circulate functions between apps, the system, and its services: the Agentic API, in alpha, relies on Gemini Nano 4 to execute functions locally on smartphones; AppFunctions API, in beta, serves as a common layer for tool calling; and the Agent-to-UI interface, still in alpha, allows an agent to display interactive components instead of a simple text block. In the same move, the Connected Apps Platform aims to bring applications closer to Search and Gemini, with finer control for publishers. The message is clear: Google no longer wants to just provide models but the entire plumbing around them.

Key points

  • More than 40 compatible agents in Android Studio Rabbit 2.
  • Google announces a 77% gain on certain UI tasks in CLI.
  • Gemini 4 Carbon ranks first on Artificial Analysis, v5.0.
  • More than 50 integrations are already active in Gemini.
  • XREAL AURA is sold for 1,099 dollars in the United States.
  • Announcement made on October 28, 2026, at the Android Dev Summit.

By the numbers

  • 40+ agents — compatible with Android Studio via BYOA, according to Google.
  • 77 % — speed gain on certain UI tasks, Google internal tests.
  • 30+ Android Skills — documentation modules provided to agents.
  • 50+ integrations — already available in the Gemini app, according to Google.
  • 1,099 dollars — launch price of XREAL AURA in North America and South Korea.

To read