Google Expands Android Developer Tooling with CLI Device Streaming and New AI Agent Skills

As the landscape of mobile development continues to evolve with the rapid adoption of artificial intelligence, Google is introducing a suite of powerful updates designed to give Android developers greater flexibility and efficiency. Whether engineers prefer working with custom command-line interfaces, advanced large language models, specialized agents, or traditional development environments, the overarching goal remains centered on helping developers build beautiful, high-quality Android applications. To support this vision, Google has announced major enhancements to its Android CLI command-line tooling, bringing remote real-device access directly to the terminal, alongside an expanding repository of grounded Android skills engineered to bridge the gap between model training cutoffs and modern platform standards.

The centerpiece of these updates is the expansion of the Android CLI, a command-line interface tool created specifically to streamline Android development workflows. Built to support virtually any AI agent or external tool, the CLI empowers developers and automated systems alike to build applications with greater efficiency. Through integration with the newly announced android-cli skill, artificial intelligence agents can leverage the command-line utility to handle a comprehensive range of project responsibilities. These include creating, building, testing, and managing complex Android codebases, assisting with the initial setup of development environments, spinning up and executing emulators, and running automated test suites seamlessly from the terminal.

Expanding upon the capabilities of command-line and agent-driven development, Google has integrated Android Device Streaming directly into the Android CLI framework. Rigorous testing on real physical devices has always been a critical step in mobile development to uncover hardware- and operating system-specific issues that emulators might occasionally miss. However, developers and automated agents frequently lack physical access to the diverse array of devices currently available on the market. By incorporating Android Device Streaming into the CLI, Google allows developers and their AI agents to tap into a vast cloud-based inventory of real physical devices remotely from anywhere.

This integration establishes a secure connection to physical hardware using ADB over SSL, operating much like a device plugged directly into a workstation via USB. Through this secure terminal bridge, an AI agent can execute a wide variety of tasks headlessly. Agents can spin up specific physical hardware, deploy fresh application builds, collect detailed logs and diagnostic traces, and even capture screenshots directly through the command line. By linking a project and instructing an agent to query available remote devices, developers can incorporate physical hardware testing directly into automated development loops, ensuring that applications perform reliably across diverse form factors and device specifications before reaching end users.

To make AI agents truly effective in these complex scenarios, developers must overcome a fundamental limitation inherent to large language models: the gap between a model’s static training data and the rapidly changing standards, APIs, and best practices of modern platform development. To address this challenge, Google is continuing to expand its Android skills repository, offering a robust collection of structured instructions contained within standardized files that ground AI agents using official documentation from developer.android.com. Rather than relying on outdated training cutoffs or generalized assumptions, these skills inject precise, up-to-date data, official API references, code samples, and authoritative architectural patterns directly into the context window of the developer’s chosen AI agent.

With more than twenty specialized skills currently available in the repository, developers can equip their agents to tackle sophisticated engineering challenges with confidence. The engineering philosophy behind these skills is rooted in rigorous evaluation and maintenance, ensuring that agents receive reliable guidance tailored to the platform’s lifecycle. Furthermore, these Android skills are designed to be entirely environment-agnostic. Whether an engineer is writing code inside Android Studio, utilizing specialized development environments like Antigravity, or pairing with popular third-party coding agents such as Claude and Codex, the underlying Android skills adapt smoothly across the entire development setup, allowing teams to maintain consistency regardless of their preferred toolchain.

Among the latest additions to the repository is a specialized skill focused on a unique segment of the ecosystem: the Wear Compose Material 3 skill. Developing applications for Wear OS presents distinct design and technical hurdles that differ significantly from standard mobile development. Engineers must account for round viewports, rotary input mechanics, power-efficient ambient display modes, and specialized layout containers such as the TransformingLazyColumn, AppScaffold, and ScreenScaffolds. Without explicit, structured guidance, general-purpose large language models often struggle to understand these specialized UI patterns and architectural requirements, frequently resulting in non-compliant code or missed performance optimizations.

The introduction of the Wear Compose Material 3 skill directly addresses this knowledge gap by providing agents with the exact framework knowledge required to build polished, performant watch applications. Early enterprise adopters are already reporting substantial productivity gains after integrating the skill into their workflows. For instance, the engineering team at FotMob utilized the new skill to modernize their existing Wear Material 3 application, successfully migrating multiple legacy lists to the modern TransformingLazyColumn component complete with ScreenScaffold content padding, custom ListHeader titles, SurfaceTransformation effects on cards and buttons, updated theme typography, and comprehensive Wear previews.

The results of utilizing the skill proved transformative for the FotMob team. The resulting code compiled successfully on the first try and was rigorously verified on emulators for smooth scrolling, rotary input response, edge morphing, and right-to-left layout support. Crucially, the process allowed the team to safely eliminate their legacy wrapper code alongside piles of custom rotary and focus boilerplate. According to the team, the specialized skill even caught subtle architectural mistakes that the underlying model would have otherwise missed, such as failing to forward the essential content padding from the ScreenScaffold down into the list, and correctly applying theme typography instead of falling back to hardcoded sp values. Reflecting on the efficiency achieved during the migration, Roy Solberg, Android Tech Lead at FotMob, noted that a single skill utilized over the course of a single afternoon successfully enabled the migration of eight distinct lists while wiping out an accumulation of custom rotary interaction code.