Google pushes Gemini everywhere

Google is putting Gemini into the plumbing of its products, from search-style retrieval tools for developers to writing tools in Docs and agent software for businesses. (blog.google) One piece is Gemini Embedding 2, which Google introduced in public preview on March 10, 2026, and now documents as the latest embedding model in the Gemini application programming interface. Embeddings turn text, images, audio, video, and documents into numeric vectors so software can find related material without relying on exact keywords. (blog.google, ai.google.dev) Google says Gemini Embedding 2 works across more than 100 languages and accepts text, images, audio, video, and Portable Document Format files in one shared vector space. In Vertex AI, the model produces vectors up to 3,072 dimensions and supports up to 8,192 input tokens, with settings to shrink output size for cheaper retrieval jobs. (ai.google.dev, docs.cloud.google.com) In Google Docs, the company announced on April 22 a redesigned Gemini writing panel that can draft, rewrite, and format documents using context from Drive, Gmail, Chat, and the web. Google also added “Match writing style” and “Match doc format,” with English first and Spanish, Japanese, Brazilian Portuguese, French, German, Italian, and Korean scheduled to follow. (workspaceupdates.googleblog.com) For corporate customers, Google used Cloud Next on April 22 to turn Vertex AI into the Gemini Enterprise Agent Platform. Google said the new platform combines model building with agent integration, orchestration, security, DevOps, and governance tools, and that future Vertex AI roadmap updates will ship through the new platform rather than as a standalone service. (cloud.google.com, blog.google) Google’s pitch is that companies want agents that do more than answer prompts. Computerworld reported that Gemini Enterprise is adding collaborative “Projects,” a Canvas tool for co-editing Google Docs, Slides, and Microsoft 365 files, an Inbox for tracking agent activity, and longer-running agents that can handle multi-step workflows such as financial reconciliations. (computerworld.com) Google is also stressing controls as it widens access. Computerworld reported that Agent Designer, a no-code builder for schedule- or trigger-based agents, includes a visual workflow view and human approval checkpoints, while Google says the platform is built to keep agents tied into enterprise security and information technology operations. (computerworld.com, cloud.google.com) The company is not limiting Gemini to Google-owned surfaces. At Cloud Next, Google Cloud chief Thomas Kurian said a more personalized Siri powered by Gemini is coming later in 2026, according to reports from MacRumors and AppleInsider citing Google’s remarks. (macrumors.com, appleinsider.com) Taken together, the announcements show Google moving Gemini from a model people try into software other products call, employees use inside daily workflows, and developers build into production systems. The common thread across the updates is less about a single chatbot and more about making Gemini the default layer underneath search, writing, and agent work. (blog.google, cloud.google.com, workspaceupdates.googleblog.com)