Android CLI: Build Android apps 3x faster using any agent
TL;DR Highlight
Google has released Android CLI and Android Skills for AI agent-based Android development, achieving a 70% reduction in LLM token usage and a 3x speed improvement in internal experiments.
Who Should Read
Developers utilizing AI agents such as Claude Code, Codex, and Gemini CLI to develop Android apps, or developers looking to set up Android build pipelines in terminal/CI environments without Android Studio.
Core Mechanics
- Google has unveiled 'Android CLI,' a collection of Android-specific CLI tools designed for use with AI agents. It's designed to work with third-party agents like Claude Code, Codex, and Gemini CLI.
- Internal experiments show that using Android CLI reduces LLM token usage by more than 70% and speeds up task completion by 3x compared to using only standard tools. This is largely due to the reduction in tokens wasted on agents exploring SDK setup or project creation.
- `android sdk install` allows selective installation of only the necessary SDK components, keeping the development environment lightweight. Creating projects with `android create` based on official templates applies recommended architectures and best practices from the start.
- `android emulator` allows you to create virtual devices and `android run` to deploy apps, making it easier for agents to automate the build-deploy cycle. You can also keep the CLI up-to-date with `android update`.
- The 'Android Skills' GitHub repository has also been released to inject Android-specific knowledge into LLMs. Skills are modular command sets in SKILL.md format, designed to be triggered automatically based on prompt content.
- Currently available Skills include Navigation 3 setup and migration, implementing edge-to-edge UI support, migrating to AGP (Android Gradle Plugin) 9, transitioning from XML to Compose, and analyzing R8 (code shrinking/obfuscation tool) settings.
- Android CLI is designed not only as a tool for agents but also for script-based workflows such as CI/CD pipelines and maintenance automation. Switching to Android Studio is also possible at any time.
Evidence
- There was a lot of agreement with a comment complaining about 'Why is a proper Android build interface coming out just to please AI chatbots?' It revealed a bitterness that something developers have been requesting for years is only being resolved because of AI agents.
- From the perspective of Apple/iOS developers, there was an opinion that 'I'm developing for macOS/iOS without Xcode and it's too difficult to keep up with the changes. I wish there was something like this for iOS.' Specific cases were also mentioned, such as the difficulty of implementing certain features like AUv3 (Audio Unit plugin format) in a terminal-based environment.
- There were complaints about the Privacy Policy clause stating that Google collects Android CLI usage data (commands, subcommands, flags). While it can be disabled with the `--no-metrics` flag, the fact that it cannot be set via an environment variable was pointed out, and there were reactions like 'Doesn't Google already have enough data?'
- There was also a cynical comment stating that even with AI and large companies with sufficient resources, they will continue to use WebView apps or cross-platform wrappers. This reflected a skeptical view of whether the decision to choose native app development will change even if the tools improve.
- While easily deploying apps is good, there was also a request to 'allow publishing apps to the Play Store without a personal ID.' This revealed that the ecosystem's barriers to entry (identity verification requirements) remain a source of dissatisfaction, as much as tool improvements.
How to Apply
- When starting a new Android project with an AI agent like Claude Code or Codex, creating a project based on an official template with the `android create` command can reduce the tokens wasted on the agent setting up the architecture and allow you to focus directly on implementing features.
- If setting up an Android build environment in a CI/CD pipeline is complex, writing a script to install only the necessary components with `android sdk install` can save time and space.
- When assigning tasks such as Navigation 3 migration or Compose transition to an agent, adding the SKILL.md file from the Android Skills repository to the project or setting it with the `android skills` command can encourage the agent to follow the latest recommended practices instead of outdated patterns.
- If you want the agent to automate UI testing as well, integrate the flow of creating a virtual device with `android emulator` and deploying with `android run` into the agent workflow to run the entire build-deploy-test cycle without human intervention.
Code Example
# Android CLI main command examples
# Install SDK components (selectively only what you need)
android sdk install
# Create a new project with the official template
android create
# Create and manage virtual devices (emulators)
android emulator
# Build the app and deploy to the device
android run
# Update the CLI to the latest version
android update
# View available Skills list and settings
android skills
# Disable usage data collection
android <command> --no-metricsTerminology
Related Papers
Show HN: adamsreview – better multi-agent PR reviews for Claude Code
Claude Code에서 최대 7개의 병렬 서브 에이전트가 각각 다른 관점으로 PR을 리뷰하고, 자동 수정까지 해주는 오픈소스 플러그인이다. 기존 /review나 CodeRabbit보다 실제 버그를 더 많이 잡는다고 주장하지만 커뮤니티에서는 복잡도와 실효성에 대한 회의론도 나왔다.
How Fast Does Claude, Acting as a User Space IP Stack, Respond to Pings?
Claude Code에게 IP 패킷을 직접 파싱하고 ICMP echo reply를 구성하도록 시켜서 실제로 ping에 응답하게 만든 실험으로, 'Markdown이 곧 코드이고 LLM이 프로세서'라는 아이디어를 네트워크 스택 수준까지 밀어붙인 재미있는 사례다.
Show HN: Git for AI Agents
AI 코딩 에이전트(Claude Code 등)가 수행한 모든 툴 호출을 자동으로 추적하고, 어떤 프롬프트가 어느 코드 줄을 작성했는지 blame까지 가능한 버전 관리 도구다.
Principles for agent-native CLIs
AI 에이전트가 CLI 도구를 더 잘 사용할 수 있도록 설계하는 원칙들을 정리한 글로, 에이전트가 CLI를 도구로 활용하는 빈도가 높아지면서 이 설계 방식이 실용적으로 중요해지고 있다.
Agent-harness-kit scaffolding for multi-agent workflows (MCP, provider-agnostic)
여러 AI 에이전트가 서로 역할을 나눠 협업할 수 있도록 조율하는 scaffolding 도구로, Vite처럼 설정 없이 빠르게 멀티 에이전트 파이프라인을 구성할 수 있다.
Show HN: Tilde.run – Agent sandbox with a transactional, versioned filesystem
AI 에이전트가 실제 프로덕션 데이터를 건드려도 롤백할 수 있는 격리된 샌드박스 환경을 제공하는 도구로, GitHub/S3/Google Drive를 하나의 버전 관리 파일시스템으로 묶어준다.