Building agents with the ADK and the new Interactions API

Google has launched a beta Interactions API that provides a unified, stateful interface to both raw Gemini models and the fully managed Gemini Deep Research Agent, designed for multi‑turn, agentic workflows; it lets ADK-based agents offload context and reasoning loops to the server (via interaction IDs and asynchronous polling) and maps the Agent2Agent (A2A) protocol through an InteractionsApiTransport so existing agent ecosystems can call Gemini as a remote agent with minimal code changes. This reduces client-side boilerplate and timeouts, simplifies integration for multi-agent systems, and gives developers immediate access to server‑side context management and Deep Research capabilities. For investors and allocators, the API lowers engineering friction and could accelerate enterprise adoption of Google’s AI platform—potentially increasing demand for Google’s model and cloud services—although some A2A features (push notifications, extensions and callbacks) are not yet supported in this initial release.

Analysis

Google has launched a beta Interactions API that provides a unified, stateful interface to both raw Gemini models and the managed Gemini Deep Research Agent, explicitly targeting multi-turn, agentic workflows. The API introduces server-side interaction IDs and asynchronous polling so ADK agents can offload the reasoning loop, avoid client-side timeouts and maintain lighter-weight pointers to server state; example model references include gemini-3-pro-preview, gemini-2.5-flash and agent="deep-research-pro-preview-12-2025".

The Interactions API is presented as both an alternative to the existing generateContent endpoint and as a primitive for ADK frameworks, while the InteractionsApiTransport maps the Agent2Agent (A2A) protocol (e.g., SendMessage→create Task, TaskStatus→Interaction Status) so existing multi-agent systems can treat a Google-hosted agent as a remote A2A peer. This reduces integration boilerplate and preserves the developer experience via AgentCard and client_factory patterns, enabling an inner-loop/outer-loop design where heavy context management is handled server-side.

From a go-to-market standpoint the release lowers engineering friction and could accelerate enterprise experimentation with Google’s model and cloud services, consistent with a mildly positive sentiment score of 0.35. Material caveats include beta status and missing A2A features (push notifications, extensions, callbacks) that limit some integrations; timelines for GA, enterprise adoption, and monetization remain key uncertainties to monitor.

Key Decisions for Investors

• Consider a modest overweight in GOOGL/GOOG on the thesis that the Interactions API beta reduces engineering friction and could expand demand for Google’s AI and cloud services, while sizing exposure to reflect beta-stage execution risk
• Monitor adoption and product milestones closely — specifically ADK/A2A integration case studies, developer engagement metrics, GA timing, and delivery of A2A features (push notifications, extensions, callbacks) — and re-rate the position on clear enterprise wins or feature parity
• Maintain downside protection via conservative position sizing or option hedges until the Interactions API exits beta and demonstrates a clear monetization path and enterprise traction