Microsoft released an agentic web interface on Monday that can perform complex tasks on the web. Announced at the company’s Build 2025 conference, the new tool is named Magentic-UI, and the company is pitching it as a “human-centred web agent.” The Redmond-based tech giant claims that, unlike other computer use agents, this one collaborates with the user to complete a project and transparently shows the actions it takes in a separate panel. Since Magentic-UI is an open-source agent, the company has made it available on GitHub as well as its Azure AI Foundry Labs.
Magentic-UI Arrives as Microsoft’s First Computer Use Agent
In its research blog, the tech giant announced the release of the Magentic-UI agentic web interface and detailed its workings. Essentially, it is a computer-focused network of agents, similar to OpenAI’s Operator or Anthropic’s Computer Use. These agents can browse and navigate through a virtual machine and access apps such as web browsers to execute actions. These can be used to perform tasks such as reserving a table at a restaurant or purchasing a product online.
The tech giant states that Magentic-UI can browse the web, write and execute Python and shell code, and analyse files. Highlighting the “human-centric” moniker, Microsoft said that the AI agent allows users to modify its plans, lets users pause the system and give feedback in natural language, or even take control of the browser back.
Magentic-UI’s interface
Photo Credit: Microsoft
Additionally, the agentic network will seek user approval before taking any action that is deemed irreversible. Such actions include placing an order from a restaurant, adding products to the cart, and more. Users can also modify the command at this time. Further, Magentic-UI can also learn from and remember previous tasks and use that information to complete future tasks.
Coming to the architecture, it is built on the company’s Magentic-One agentic system. Microsoft says Magentic-UI is a combination of four different AI agents, working together to complete a task. The interface includes an Orchestrator or the lead agent that makes the task plan alongside the user. It also incorporates feedback and delegates sub-tasks to other agents. Then there is a WebSurfer agent that is equipped with a web browser. It can click, type, scroll, and visit pages in multiple rounds to complete a task.
Then there is a Coder agent that comes with a Docker code-execution container. It can write and execute Python and shell commands. And finally, the FireSurfer is also equipped with a Docker code-execution container and file conversion tools.
Those interested in downloading and using the AI agent in a local setting can access it via GitHub. Microsoft is offering the research preview of the agent with a permissive MIT licence. For those who prefer to access the cloud-based version of Magentic-UI, the company is hosting it on its Azure AI Foundry Labs.
