Google is expanding Gemini from a chatbot into a work agent that can plan tasks and act across multiple applications. At the same time, the company is testing Google AI Edge Foresight, a meeting notetaker that processes conversations entirely on your Mac. Both developments matter because they represent contrasting approaches to Artificial Intelligence (AI): a broadly connected cloud service and a local tool that works without an internet connection.
From chatbot to work agent
An AI agent is a system that goes beyond answering questions by breaking a goal into steps and using applications or data sources to complete it. According to Google, the new Gemini agent will be available through one central interface and can continue working in the background. It is launching inside Gemini Enterprise, Google’s business offering, and is initially limited to a private preview with selected participants.
The agent works directly with Gmail, Drive, Docs, Sheets, and Calendar. According to The Verge’s lead report, it can also interact with services including Microsoft 365 and Slack. Because it runs in the cloud, meaning on remote computing infrastructure rather than only on your device, it can retain the same work context across phones, computers, and web browsers.
Google presents the agent as a central entry point for knowledge work, media creation, and writing or running code. A MarkTechPost summary emphasizes that these capabilities are available through a single prompt box. For everyday users, the more relevant point is not the underlying technical interface but the prospect of starting different kinds of work without moving among several separate AI tools.
The main difference from a conventional chatbot is the type of request you can make. Instead of specifying every individual step, you should be able to give the agent an objective. Gemini then plans the process, accesses connected systems, and selects what Google considers the most suitable AI model; users can also choose a model themselves, initially including Anthropic’s Claude.
What it can do at work
Tasks mentioned by Google include scheduling meetings and booking appointments or travel. One concrete workplace example is arranging a meeting while accounting for calendars, time zones, and required approvals. Rather than merely suggesting a suitable time, the agent is intended to complete the action on the user’s behalf.
For larger assignments, you can attach files, folders, and projects organized around a particular workflow. The agent can also connect to Google Workspace, Microsoft 365, Slack, Jira, Confluence, and several business data systems. A team could therefore provide existing project files and internal information for a task instead of copying every relevant passage into a chat window.
Gemini can delegate work to specialized subagents, meaning AI helpers focused on particular parts of a larger assignment. A tasks inbox is designed to show which steps are underway, which special capabilities have been loaded, and how far the assignment has progressed. This provides more visibility than a chat response that simply appears after an unseen process.
Google takes the digital coworker concept further by giving the agent its own Workspace identity. It can receive an email address, respond when tagged or contacted, and join group conversations. According to the company, its actions produce an audit trail, a traceable record that attributes the work to the agent rather than to a human employee.
TechCrunch reports that Gemini has more than 1 billion monthly active users and that nearly 90% of Fortune 100 companies use Gemini Enterprise. These figures come from Google and were not independently verified in the supplied reports. The company is focusing on businesses first, saying this will help it address security, performance, and operating at scale before a broader consumer rollout.
Local notes keep data on the device
Google AI Edge Foresight takes a markedly different approach. The experimental Mac app is free to use and can transcribe meetings and audio files entirely offline. An on-device model, meaning AI that runs directly on your computer instead of in the cloud, processes audio, notes, and connected files without transferring them off the device, according to Google.
The app places short notes written by the user alongside expanded notes generated by AI. During a meeting, you can record brief bullet points, and Foresight uses the transcript to turn them into more polished notes. A built-in assistant can also answer questions about the conversation and refer to local files for additional context.
A practical example is an in-person or professional meeting in a location without a reliable internet connection. Foresight can still transcribe the conversation, and you can search the notes later while traveling offline. The account of its offline processing relies on Google’s claims, however, and the supplied sources do not cite an independent technical assessment.
Supported material includes PDFs, Google Docs, Microsoft Office formats, plain text, Markdown, and saved web bookmarks. These items can form a local knowledge base that the assistant consults when answering questions. According to TechCrunch’s Foresight report, transcription uses EmbeddingGemma 2 with 740 million parameters, while a Gemma 4-powered assistant handles questions.
The limitation is substantial: Foresight is currently optimized for Macs with Apple Silicon. The reports do not mention a Windows version, supported languages, availability in Switzerland, or transcription quality across different dialects. Local processing may be attractive to Swiss users and organizations, but claims about Swiss German performance or compliance with particular privacy requirements would be speculative without further information.
Pros and Cons of the new Gemini tools
Pros:
- Less application switching – The work agent brings tasks from Google Workspace, Microsoft 365, Slack, and other systems into one interface.
- Action instead of answers – Gemini is intended to plan objectives, delegate subtasks, and complete actions such as scheduling meetings.
- Traceable work – A tasks inbox and audit trail are designed to expose the agent’s progress and recorded actions.
- Local privacy – According to Google, Foresight processes conversation data and files offline on the Mac rather than sending them to the cloud.
Cons:
- Limited access – The work agent is initially available only to enterprise customers in a private preview, and the sources provide no price.
- Broad data access – An agent connected to calendars, email, files, and internal systems receives extensive context, making permissions and control especially relevant.
- Unverified reliability – The reports contain no independent testing of complex agent tasks or Foresight’s transcription accuracy.
- Narrow device support – The local notetaker is optimized for Apple Silicon Macs and is not a generally available option for every workplace.
What this means for you
If you are getting started, Foresight is the more contained concept: one application performs a clearly defined job and, according to the provider, keeps processing local. A sensible first step would be to try it with a non-sensitive meeting or an existing audio file and compare the transcript with your own notes. The sources do not establish whether it will reliably recognize your language, specialist vocabulary, or dialect.
With the work agent, you should begin with tightly scoped and verifiable assignments if your organization gains access. Scheduling a meeting or gathering existing information is easier to review than a request that makes broad changes across several systems. Since Gemini is supposed to record its actions, you can compare the visible process with the objective you originally assigned.
Advanced users can gain more by deliberately connecting suitable files, projects, and internal tools, then dividing complex objectives into clearly bounded responsibilities. Model selection and specialized subagents provide additional flexibility. They also increase the need for oversight, because placing actions behind one convenient interface does not make every result correct.
Important practical questions remain unanswered for Swiss organizations. The sources provide no pricing, firm availability details for Switzerland, language coverage, or information about where cloud data is processed. Foresight’s local design may reduce the transfer of meeting content according to the provider, but it does not replace an organization’s review of devices, access permissions, and internal requirements.
Together, the two products illustrate two credible directions for workplace AI: the Gemini agent connects as many systems as possible and performs work, while Foresight keeps potentially sensitive material on the device. Their immediate value lies in reducing manual handoffs and making meeting information easier to retrieve. The unresolved issue is how reliably these tools perform outside controlled demonstrations and how much practical control users retain when agents receive broad assignments.
Sources
- Google is launching a one-stop Gemini agent for your work tasks – The Verge, 2026-10-08
- Google Cloud Launches Gemini Agent, One Universal Agent for Enterprise Work – MarkTechPost, 2026-10-09
- Google brings agentic AI to Gemini, starting with businesses – TechCrunch, 2026-10-08
- Google’s AI note-taking app transcribes your meetings completely offline – The Verge, 2026-10-08
- Google releases a new local-first Granola competitor – TechCrunch, 2026-10-08
Image: freestocks.org via Pexels