Google Unveils Major Suite of Developer Tools and APIs to Build Agentic, Multi-App Experiences on Android

As the technology landscape shifts into the intelligence era, users are no longer simply tapping through individual applications to accomplish routine digital chores. Instead, modern mobile users are increasingly expressing high-level, complex goals and expecting intelligent AI agents—such as Gemini—to seamlessly complete multi-step workflows. Often, these workflows span across multiple discrete applications on a user’s device, requiring a deeply integrated software ecosystem. To help developers meet users where they are and capitalize on this paradigm shift, Google has introduced a comprehensive suite of new developer tools, frameworks, and APIs designed to build agentic, highly intelligent experiences directly into Android applications.

The latest wave of capabilities was previewed during the Android Developer Summit keynote, offering a glimpse into the future of ambient and proactive computing. To support this transition, Google has rolled out deep-dive resources, including ten new on-demand developer sessions dedicated entirely to engineering intelligent applications. These resources aim to guide software engineers through the complexities of integrating local on-device models, hybrid cloud-device architectures, and standardized tool-calling frameworks into their existing mobile products.

Connect with the Android Intelligence System

At the heart of Google’s new developer initiative is the Android intelligence system, a robust foundational layer engineered to protect user privacy and data security while safely connecting advanced AI agents with installed applications. This system spans the entire Android device ecosystem, supporting smartphones, wearables, extended reality (XR) glasses, television sets, and automotive interfaces. It is designed to deliver both prompted assistance when a user explicitly asks for help and proactive assistance tailored to context and user habits.

To give developers precise control over how their applications interact with this overarching intelligence architecture, Google is providing standardized APIs and developer tools built on the Model Context Protocol (MCP). Android MCP, which is powered by the new AppFunctions Jetpack library, serves as a unified tool-calling framework. This framework allows third-party applications to establish secure, direct communication channels with on-device AI agents.

The ecosystem distinguishes between remote MCP servers, which offer broad cross-platform reach, and Android MCP, which acts as a native local solution. Google recommends Android MCP as the primary integration path for user journeys that involve end-to-end encryption, rely on sensitive local device data, require access to device sensors and hardware, need to execute entirely locally, or operate in environments where no cloud server endpoint is available. Depending on specific product requirements, developers can choose to integrate their applications using remote MCP, Android MCP, or a flexible combination of both.

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

The evolution of on-device machine learning continues to accelerate with new capabilities introduced in Gemini Nano 4, including advanced function calling. By expanding the ML Kit GenAI APIs, Google is enabling developers to construct sophisticated agentic workflows that operate entirely on-device without needing round-trips to the cloud. Developers are provided with direct building blocks through ML Kit’s direct model access via the Prompt API, alongside the newly introduced Agentic API.

This Agentic API is designed to streamline development by automatically executing functions, saving conversation history, managing complex orchestration loops, and enforcing automatic safety guardrails. Developers looking to explore these features can examine the Jetpacker code sample available on GitHub, which demonstrates how to unlock powerful agentic capabilities within standard mobile applications using ML Kit GenAI APIs.

In addition to the Agentic API, Google is introducing a suite of complementary on-device capabilities designed to enhance app responsiveness and intelligence. For application workflows that require a balanced mix of cloud processing and local inference, Firebase AI Logic offers a straightforward hybrid solution. This setup allows developers to leverage advanced, resource-intensive cloud models like Gemini 4 while simultaneously utilizing fast, local inference through Gemini Nano whenever appropriate. To help production environments remain secure and cost-effective, developers can implement Firebase App Check, utilize new server prompt templates, and enforce strict Firebase spend caps.

Furthermore, addressing how AI agents present information to users, Google has introduced the Jetpack Compose A2UI renderer in alpha. Showcased during the keynote demonstration via Jetpacker’s live booking assistant, this renderer allows AI agents to effectively "speak UI." Agents can stream rich, interactive native Compose user interfaces directly derived from a lightweight JSON protocol, bridging the gap between conversational AI and traditional mobile interfaces.

Reach Across Gemini and Search with Connected Apps Platform

As user behavior across Google surfaces transitions from simply asking informational questions to actively executing tasks, developers need efficient ways to capture high-intent traffic. To address this, Google has introduced the Connected Apps Platform. This platform allows developers to reach engaged users across both Gemini and Search by combining Android MCP for native, on-device execution with remote MCP servers for expansive cross-platform reach.

Developers can begin building and testing their integrations locally today using Gemini Apps Developer Mode. For teams preparing for production deployment, Google has opened a preview waitlist. Initially, preview capacity is being limited to selected, production-ready applications within supported categories, with broader self-service access scheduled to roll out at a later date. Comprehensive instructional resources are also available to walk development teams through the process of building, testing, and publishing their first connected applications.

Start Building Intelligent Apps with the Latest Updates

The expansive collection of tools, libraries, and frameworks released by Google gives developers everything they need to start shaping the next generation of mobile software. Engineering teams can immediately begin creating on-device agents using ML Kit agentic workflows paired with direct model access via new Android skills. Developers testing preview models are encouraged to opt into the AIcore Developer Preview program to evaluate upcoming capabilities.

By combining ML Kit APIs with custom machine learning models running through LiteRT in Google Play Services, developers can embed sophisticated predictive and generative features directly into their codebases. Cloud-based inference with Gemini models can be managed seamlessly through Firebase AI Logic, while native user interfaces can be dynamically streamed from JSON using the Jetpack Compose A2UI renderer.

Additionally, integrating applications with the Android intelligence system allows for robust prototyping with Android MCP using AppFunctions skills and sample agents. Developers can also sign up for the Connected Apps Platform developer preview to secure early access for production readiness. With these foundational systems now in place, the path is cleared for developers to design the next wave of intelligent, agent-driven mobile experiences.