Agentic interaction using AppFunctions
agents
| Source: Dev.to | Original article
Google has unveiled **AppFunctions**, a new Android‑level API that lets generative‑AI agents invoke app capabilities directly, bypassing screen‑scraping or accessibility hacks. The feature ships as part of Android 16 and a matching Jetpack library, enabling developers to declare discrete functions—such as “send‑money”, “book‑flight” or “fetch‑calendar events”—that the OS can expose to AI assistants like Gemini, Claude or third‑party agentic bots.
The move responds to the rapid expansion of “agentic interaction” on mobile, where AI agents orchestrate multi‑step workflows across apps. By offering a structured, on‑device contract for function calls, AppFunctions promises lower latency, stronger privacy (no need to transmit raw UI data) and more reliable execution than the brittle automation scripts that have dominated the space. Google positions the API as the mobile analogue of the Model Context Protocol (MCP) used in server‑side tool calling, a pattern we covered last week in the context of Amazon SageMaker’s serverless model customisation and UnionPay’s open payment protocol.
Why it matters is twofold. First, it lowers the barrier for app owners to become AI‑ready; the library can auto‑generate the necessary manifest entries without code changes, meaning even legacy apps can be queried by agents. Second, it gives Google a foothold in the emerging ecosystem of agentic tooling, potentially steering standards for how on‑device AI interacts with third‑party services. The approach also aligns with broader industry pushes for open, trustworthy AI interfaces, echoing the APEX standard for agentic trading and the Holos multi‑agent web framework.
Looking ahead, developers will need to adopt the Jetpack AppFunctions SDK and publish function schemas to the Play Store’s AI catalogue. Watch for the first wave of Gemini‑powered Android experiences in the coming months, and for competing platforms—Apple’s rumored “Intents for AI” and third‑party SDKs—to either adopt Google’s schema or propose alternatives. The speed at which app ecosystems embrace these contracts will determine whether agentic AI becomes a seamless mobile layer or remains a niche experiment.
Sources
Back to AIPULSEN