Microsoft unveils Magentic-UI, a human-centred AI agent for the web: How it works

Microsoft unveiled Magentic-UI, its first computer use agent, at Build 2025. This human-centred web agent enhances real-time collaboration, ensuring user control and transparency through a visible task panel. It supports various online tasks while prioritizing user consent for irreversible actions.

Livemint
Published20 May 2025, 07:01 PM IST
Microsoft has introduced its first-ever computer use agent, Magentic-UI, at the Build 2025 developer conference.
Microsoft has introduced its first-ever computer use agent, Magentic-UI, at the Build 2025 developer conference.

Microsoft has introduced its first-ever computer use agent, Magentic-UI, at the Build 2025 developer conference. Described as a "human-centred web agent", the new interface is designed to collaborate with users in performing complex online tasks while maintaining full transparency of its actions.

Unlike traditional AI agents, Magentic-UI operates with the user at the centre of its processes. It enables real-time project collaboration, giving users the ability to modify plans, pause execution, provide feedback in natural language, or even resume direct control. A key feature of this system is its visible task panel, which shows all agent actions step-by-step—aiming to build user trust and promote transparency.

Also Read | Elon Musk’s Grok AI now runs on Microsoft Azure despite OpenAI lawsuit

Microsoft is positioning Magentic-UI as part of its broader vision of agentic computing, similar in function to tools such as OpenAI’s Operator or Anthropic’s Computer Use. The platform supports web browsing, code execution in Python and shell environments, and file analysis. It is capable of performing tasks such as making restaurant reservations or completing online purchases—all while asking for explicit user consent before executing any irreversible action, like placing an order or adding items to a shopping cart.

The system’s architecture is built on Magentic-One, Microsoft’s proprietary agentic framework, and consists of four distinct AI agents working in unison:

  • The Orchestrator: Acts as the lead planner, co-developing task strategies with the user and delegating responsibilities to sub-agents.
  • WebSurfer: Equipped with a built-in browser, this agent can navigate websites, click, type, scroll, and complete multi-step processes online.
  • Coder: This agent can write and execute Python and shell commands within a Docker-based container.
  • FireSurfer: Also operating in a Docker environment, this agent handles file conversions and code executions, expanding the interface’s versatility.
    In a move aimed at the developer and open-source community, Microsoft has made Magentic-UI freely available on GitHub under the permissive MIT licence. A cloud-based version is also accessible via Azure AI Foundry Labs.

Catch all the Technology News and Updates on Live Mint. Download The Mint News App to get Daily Market Updates & Live Business News.

Business NewsTechnologyNewsMicrosoft unveils Magentic-UI, a human-centred AI agent for the web: How it works
MoreLess