Google is putting AI notes on the device. The company has released Google AI Edge Foresight, an offline-capable Mac app that competes with Granola and targets meeting notes without requiring a constant cloud connection. At the same time, Google Cloud introduced a broader Gemini agent for enterprise work, giving Google’s AI push two distinct directions: private local assistance and connected workplace automation.
Google AI Edge Foresight runs on Apple Silicon and uses the on-device EmbeddingGemma 2 model, which has 740 million parameters. Google says the app “is optimized for Apple Silicon and can completely work offline,” a useful distinction in a market where AI products often treat an internet connection as part of the furniture.
The app can work with meeting notes, view transcripts, chat with an AI assistant, and upload documents to build a knowledge base. Those features place it close to the expanding group of AI note-taking tools, including products released recently by Wispr, Calendly, and Superhuman, formerly Grammarly. Superhuman also acquired Fathom, a note-taker, adding another name to a category that has started collecting companies at an impressive rate.
Local notes first, cloud features later
Google’s earlier dictation app remained experimental and never reached mainstream usage. Google AI Edge Foresight gives the company another attempt at a focused productivity tool, but its local design separates it from assistants built around cloud models and constant service access.
Google could release a consumer version that works across video calling apps through Gemini apps relying on cloud models. That would extend the product beyond the Mac app’s current local-first approach, although the two versions would serve different needs: one keeps processing on the device, while the other connects meeting activity to Gemini’s cloud-based capabilities.
The timing puts Google inside a crowded contest. Wispr, Calendly, and Superhuman have released their own note-takers, while Meta has released the Muse AI agent and OpenAI has launched always-on agents called dots. AI companies are no longer treating notes, dictation, and task handling as separate product categories—because apparently every meeting now needs its own software employee.
Gemini becomes a workplace operator
Google Cloud introduced the Gemini agent on October 8, 2026, as part of the Gemini at Work event. The Google Cloud announcement was published at 11:50 AM EDT, and an article about the agent was published at 2:28 PM UTC the same day.
The Gemini agent can answer questions, handle tasks, create content, and write code. It works inside Google Workspace apps including Gmail, Docs, Sheets, and Calendar, and users can also access it through Microsoft 365 and Slack.
The agent is designed to operate across apps and devices through a single interface inside the Gemini Enterprise app. Users can chat with it and assign tasks from that interface, while job-specific sub-agents can handle narrower responsibilities. Google describes the system’s workflow in plain terms: “The Gemini agent plans the work, uses tools, connects to a company’s business systems and returns finished work in documents, email and developer environments.”
The current system runs on Google’s Gemini models and Anthropic’s Claude models, with other models planned for later addition. Users can also create “coworker agents” with their own email address and limited access, giving the agent a workplace identity without handing it the keys to everything. That limitation matters; enterprise automation without access controls is just a faster route to an incident report.
Google has versions of the Gemini agent for financial services and legal work in preview, while versions for government, healthcare, and retail are coming soon. For now, the Gemini agent is available only to enterprise customers in private preview, so its capabilities remain an announcement rather than a tool most workers can use today.
Google’s two releases show the company covering both ends of the AI productivity market. Google AI Edge Foresight keeps meeting intelligence on an Apple Silicon Mac, while the Gemini agent coordinates work across applications, devices, business systems, and model providers. The strategy is broad, but the underlying question is narrow: will users trust AI with a transcript on their laptop before they trust it with work spread across the entire company?
Based on
