Google Expands Android AI Capabilities With New Agentic APIs, Model Control Frameworks, and Cross-Platform Tools

As the mobile computing landscape transitions firmly into the intelligence era, users increasingly expect their devices to do more than simply launch applications and wait for taps. Modern software experiences are shifting toward high-level goal expression, where users articulate complex intentions and rely on advanced artificial intelligence agents, such as Gemini, to orchestrate multi-step workflows across disparate applications on their devices. To address this paradigm shift and empower software creators to meet users directly within these evolving ecosystems, Google has unveiled a robust suite of new developer tools, application programming interfaces, and foundational frameworks designed for building intelligent, agentic experiences across the Android ecosystem.

Initially previewed during the Android Developer Summit keynote, these powerful additions to the Android development toolkit are now supported by comprehensive deep-dive resources, including ten new on-demand sessions tailored specifically for engineers looking to design advanced AI-driven applications. This latest release provides developers with unprecedented control over how their apps integrate with system-level intelligence while ensuring privacy, security, and performance remain paramount.

Connect with the Android Intelligence System

At the heart of this developer push is the Android intelligence system, a foundational architecture engineered to protect user privacy while seamlessly connecting intelligent AI agents with applications installed across a diverse array of Android-powered devices. Spanning smartphones, wearable devices, extended reality (XR) glasses, smart televisions, and connected automobiles, this ecosystem infrastructure delivers both prompted assistance triggered directly by the user and proactive assistance anticipated by the system.

To give developers granular control over how their software integrates with this intelligence layer, Google has introduced robust APIs and tooling rooted in the established Model Context Protocol (MCP) standard. Known as Android MCP and powered by the AppFunctions Jetpack library, this unified tool-calling framework serves as a bridge, connecting software applications with qualified AI agents residing locally on the device. Whether developers choose to work with the core system agent or deploy their own dedicated on-device, in-app agent, Android MCP facilitates smooth communication. Furthermore, for scenarios requiring cross-platform reach, remote MCP servers provide an extended operational scope.

Android MCP acts as an essential, native solution that complements remote architectures. It is the strongly recommended integration path for user journeys that demand end-to-end encryption, rely on sensitive local device data, utilize physical device sensors and hardware, require local execution without network latency, or operate in environments where a traditional server endpoint is simply unavailable. By leveraging these standards, developers can ensure their applications remain deeply integrated into the operating system while adhering to rigorous security standards.

Bring Intelligence into Your Apps: From On-Device Agent to A2UI Compose Renderer

The evolution of on-device processing power is further accelerated by new capabilities arriving in Gemini Nano version 4, particularly advanced function calling features. Google is significantly expanding the machine learning kit generative artificial intelligence application programming interfaces to allow developers to build sophisticated agentic workflows entirely on the device. These capabilities supply direct building blocks through ML Kit’s direct model access via the Prompt API, alongside a brand-new Agentic API.

This Agentic API handles numerous complex underlying tasks automatically, including executing functions, preserving conversation history, managing the orchestration loop, and enforcing automatic safety guardrails. Developers can examine real-world implementations through resources like the Jetpacker code sample available on GitHub, which provides a comprehensive look at unlocking advanced agentic capabilities directly within mobile applications.

Beyond these automated agentic interfaces, Google is rolling out a suite of complementary on-device capabilities designed to balance performance and privacy. For workflows that necessitate a hybrid approach combining cloud-based intelligence and local processing, Firebase AI Logic offers a streamlined solution. Developers can leverage advanced cloud models such as Gemini 4 while simultaneously running lightweight on-device inference using Gemini Nano whenever appropriate. To safeguard applications and manage operational costs efficiently in production environments, developers are encouraged to implement Firebase App Check, leverage newly introduced server prompt templates, and establish strict Firebase spend caps.

A particularly striking addition demonstrated during the keynote presentation is the Jetpack Compose A2UI renderer, which is now available in an alpha release. This innovative tool allows AI agents to effectively speak user interfaces by streaming rich, interactive native Jetpack Compose layouts directly derived from a JSON protocol. This capability was vividly showcased during the Jetpacker live booking assistant demonstration, pointing toward a future where user interfaces can dynamically adapt to the specific needs of an AI-driven task on the fly.

Reach Across Gemini and Search with Connected Apps Platform

Across Google’s expansive digital surfaces, user behavior is undergoing a fundamental transformation, shifting away from simply posing informational queries and moving toward executing direct, actionable tasks. To capitalize on this change and connect developers with high-intent users, Google has introduced the Connected Apps Platform. This platform bridges Android MCP for native on-device execution with remote MCP servers for broad cross-platform reach, spanning across both Gemini and Search surfaces.

Developers can begin building and testing their integrations locally today by utilizing Gemini Apps Developer Mode. As applications reach production readiness, developers can apply for the preview waitlist. Currently, preview capacity is deliberately limited to selected production-ready applications within supported categories, ensuring a high-quality experience before broader self-service access is rolled out to the wider development community. Comprehensive educational resources are also available to walk development teams through the nuances of building, testing, and publishing their very first connected applications.

Start Building Intelligent Apps with the Latest Updates

The expansive array of newly released tools provides a clear roadmap for teams looking to modernize their software. By combining ML Kit agentic workflows and direct model access with the latest Android device skills, developers can lay a strong foundation for autonomous application behavior. Participating in the AIcore Developer Preview program ensures access to the cutting-edge capabilities found in preview models, while LiteRT in Google Play Services facilitates the execution of custom machine learning models on consumer hardware.

Coupled with cloud-based inference capabilities via Firebase AI Logic, dynamic user interface streaming through the Jetpack Compose A2UI renderer, and robust prototyping using Android MCP via AppFunctions, the ecosystem provides all the necessary components for the next generation of software development. With the Connected Apps Platform developer preview opening doors to wider visibility across Google services, developers now have the comprehensive infrastructure required to transform how users interact with their mobile devices.