Google's Gemini Grows Up
Google is expanding Gemini's API to support unified tools and is privately testing a native Mac app that brings desktop intelligence—moves that promise tighter workflow automation and richer integrations for documentation and triage tools. The push beyond browser‑only models could make enterprise assistant integrations more seamless for government desktops. (testingcatalog.com) (technobezz.com)
Google posted a developer note on March 17, 2026 announcing that Gemini API calls can now include both Google-built tools and custom function calls in a single request to reduce client-side orchestration. (blog.google) The update adds explicit context "circulation" across tool calls and turns and introduces unique response IDs to let callers reference outputs from earlier tool invocations. (testingcatalog.com) Google extended Maps grounding and other built-in tools to the Gemini 3 model family, enabling location-aware reasoning without separate pre‑ and post‑processing steps. (blog.google) The Gemini developer docs also highlight a Document Understanding stack capable of processing up to 1,000‑page PDFs with multimodal analysis, which Google positions for large-scale doc automation. (ai.google.dev) Separately, Google has been rolling the Interactions API (beta) as a unified interface that moves multi‑turn state management server‑side to simplify agent workflows for developers. (devengoratela.com) Bloomberg reported on March 19, 2026 that Google has begun privately seeding a Gemini app for macOS to selected beta testers, with a Desktop Intelligence capability to tap other Mac apps and on‑screen content. (bloomberg.com) Reports say the macOS app will include image and video generation plus personalization features and is positioned to compete directly with OpenAI’s ChatGPT and Anthropic’s Claude in the desktop assistant market. (bloomberg.com)
