On Monday, Microsoft unveiled a new web interface capable of handling intricate online tasks, named Magentic-UI. This announcement was made during the company’s Build 2025 conference, where the technology was described as a “human-centred web agent.” Microsoft, headquartered in Redmond, asserts that its new tool distinguishes itself from other agents by collaborating directly with users, offering transparency about actions through a dedicated panel. The Magentic-UI is open-source, with its code available on GitHub and through Azure AI Foundry Labs.
Magentic-UI Marks Microsoft’s Entry into Computer Use Agents
In a recent blog post detailing the Magentic-UI, Microsoft described its functionality as a network of agents designed to assist users. This tool is akin to offerings from OpenAI and Anthropic, enabling users to navigate a virtual environment and execute various actions through applications like web browsers. Tasks such as making restaurant reservations or online purchases can now be streamlined with this new technology.
According to Microsoft, Magentic-UI’s capabilities extend to web browsing, writing and executing Python and shell scripts, as well as file analysis. The company emphasizes the “human-centric” aspect of the technology, which allows users to adjust plans, pause ongoing processes, provide feedback in natural language, and regain control of the browser as needed.
Interface of Magentic-UI
Photo Credit: Microsoft
Furthermore, the system prioritizes user consent before performing any irreversible actions like placing orders or adding products to a shopping cart. Users retain the ability to adjust commands at any point. In addition, Magentic-UI is designed to learn from past activities, allowing it to inform future tasks based on previous user interactions.
Regarding its design, Magentic-UI is constructed on Microsoft’s Magentic-One agentic framework, integrating four distinct AI agents working in tandem to fulfill tasks. The Orchestrator serves as the primary agent, formulating task strategies alongside the user while managing feedback and assigning sub-tasks to additional agents. The WebSurfer agent allows for navigation, enabling clicks, typing, scrolling, and multi-round web page visits.
The second agent, known as the Coder, operates within a Docker code-execution environment, capable of writing and executing commands in Python and shell formats. Completing the ensemble is the FireSurfer, equipped similarly with a Docker container and file conversion capabilities.
For those interested in integrating this AI into their local systems, Microsoft has made Magentic-UI available for download on GitHub under a permissive MIT license. Alternatively, users can access the cloud version by visiting Azure AI Foundry Labs, where the tool is hosted.
