The conversation on social media has taken a noticeable turn. Discussions about AI Agents are becoming more frequent, but the focus has shifted. People are less concerned with how well a model answers trivia questions and more interested in a practical matter: which Agent can actually get their work done?
This shift in focus is understandable. In recent weeks, major tech players have unveiled new products that push AI beyond simple chat and into the realm of active work assistance.
Tencent recently launched "Marvis," an AI assistant operating at the system level, available on Windows, macOS, and Android. It comes pre-loaded with six specialized agents that run 24/7, each handling a specific domain: files, browsers, applications, search, and computer control, ready to use upon installation.
Shortly after, OpenAI announced the integration of ChatGPT as an add-in for Microsoft PowerPoint, accessible to both free and business subscribers. This feature adds a sidebar where users can generate or edit presentations using natural language commands.
In the same week, Alphabet introduced Gemini Spark at its I/O 2026 event. This is a personal Agent designed to run continuously on a dedicated Google Cloud virtual machine, capable of reading emails, editing documents, and operating web browsers via Chrome, all without requiring constant user supervision.
From Answering Questions to Doing the Work
The ChatGPT add-in for PowerPoint allows users to command it to, for instance, "create an investor product demo, pulling project updates from last week's Outlook." It can fetch data, generate content, and handle formatting, all within the application. By connecting to services like Gmail and SharePoint, it attempts to integrate information, not just generate it. While fast at producing structured drafts, it currently lacks support for complex template and font handling.
Tencent's Marvis represents a different approach. It functions as a network of interconnected agents. A primary agent coordinates tasks, delegating to specialized agents for files, computers, apps, browsers, and search, creating a unified middleware layer for system, file, application, and cross-device control. For example, asking it to "find the Agent architecture PPT the PM sent me last week, I forgot the filename, it's on the desktop," prompts it to scan file contents for semantic understanding rather than performing a simple keyword search.
In practical tests, Marvis demonstrated an understanding of workflow. When asked to prepare materials for a review meeting, it created a pre-meeting checklist and a 90-minute agenda, breaking down tasks like having operations pull lead quality data and sales compile customer feedback. It also showed an ability to connect data across documents, analyzing a Word report and an Excel spreadsheet together to identify sales figures, regional rankings, and anomalies like duplicate entries.
However, response times are a current limitation. Simple queries can take around 30 seconds, while file analysis might require up to six minutes, with limited visibility into the intermediate processing stages.
A key part of Marvis's appeal is its user interface design. A sidebar "Office" page displays a 3D office scene where the Marvis and other agents are visualized as employees at workstations, with task completion stats and token usage shown. This playful animation makes the agent collaboration process tangible, reinforcing the "AI workhorse" concept more effectively than a plain tool interface.
The Race for the "AI Workstation"
The current competitive fervor for this space was significantly ignited by the rise of OpenClaw. Originally created by an independent developer, this open-source project gained massive traction in developer circles by demonstrating AI's ability to autonomously handle multi-step tasks like purchasing items online or migrating code. Its rapid adoption, reaching 250,000 GitHub stars in about 60 days, proved there was strong user interest in AI that "does work" rather than just "answers questions."
This evolution is evident in products like OpenAI's Codex, which started as a coding assistant but has expanded to include computer operation, browser control, and a "Goal Mode" for autonomous task completion. This pattern shows that a capable Agent, once proven in one domain, naturally expands into adjacent tasks.
This explains the strategic moves by major companies: Tencent targeting the OS level, Alphabet building a persistent 24/7 Agent, and Microsoft embedding Agent functionality into PowerPoint. They are all competing to establish the primary "AI workstation."
The Core of the AI Workstation
An AI workstation is not merely a computer with AI software or an extra chat window. It represents a new working relationship. Humans define goals, provide materials and permissions, and set success criteria. The AI then orchestrates actions across files, applications, browsers, and cloud services. The human role shifts from executor to manager, reviewer, and final decision-maker.
Impact on the Everyday User
For the average person, the value of an AI workstation lies in shifting from "I operate the software" to "I delegate a task." It removes the need to remember file locations or application workflows. The user states the objective, and the AI finds, reads, organizes, and uses tools to deliver a result. This makes it more suitable for general use than single-point tools. A PowerPoint plugin helps make a presentation; a mature AI workstation could also help prepare all the supporting materials for the meeting.
However, this new capability comes with new considerations. For an AI to work on your behalf, it requires access to more files, greater permissions, and deeper context. Mistakes can have broader consequences, affecting data, schedules, or communications. Before AI workstations become widespread, users will need to learn how to clearly define goals, set boundaries, and verify outcomes.
Ultimately, the core appeal of the AI workstation is its promise to handle routine tasks, potentially freeing users from repetitive "grunt work," which is a primary reason for its growing popularity.
Comments