Microsoft released an agent web interface on Monday that can do complex tasks on the web. Announced at the company’s build 2025 conference, the new equipment is named Magentic-UI, and the company is pitching it as a “human-focused web agent”. Redmond-based tech veteran claims that, unlike other computer use agents, it collaborates with the user to complete a project and transparently shows the tasks in a separate panel. Since Magentic-UI is an open-source agent, the company has made it available on Github as well as Azure AI Foundry Labs.
Magentic-UI comes as the first computer usage agent of Microsoft
In your research BlogTech veteran announced the release of Magentic-UI agentic web interface and expanding its functioning. Essentially, it is a computer-centric network of agents similar to the operator of Openai or the computer use of anthropic. These agents can browse and navigate through a virtual machine and execute actions to access web browser such as apps. They can be used to do tasks like burning tables or buying online products in a restaurant.
Tech giants say the magentic-UI web can browse, write and execute the python and shell code, and analyze the files. Highlighting the “human-centered” monkey, Microsoft stated that AI agents allow users to modify their plans, allow users to stop the system and respond in natural language, or even take control of the browser.
Magentic-UI interface
Photo Credit: Microsoft
Additionally, the agent network will take the approval of the user before taking any action which is considered irreversible. Such tasks include an order from a restaurant, adding products to the carts, and much more. Users can also modify the command at this time. In addition, Magentic-UI can also learn from previous tasks and remember and use that information to complete future tasks.
In the architecture, it is built on the company’s magentic-one agent system. Microsoft says that Magentic-UI is a combination of four different AI agents, which is working together to complete a task. The interface includes an orchestrator or lead agent that makes a task plan with the user. It also incorporates feedback and represents other agents to sub-functions. Then there is a websurfer agent equipped with a web browser. It can click, type, scroll and visit the page in several rounds to complete a task.
Then there is a coder agent that comes with a doctor code-performance container. It can write and execute the python and shell command. And finally, the firecorn is also equipped with a dock code-performance container and file conversion tools.
People wishing to download and use AI agent in local settings access This through Github. Microsoft is offering the research preview of the agent with an permissible MIT license. For those who like to reach the cloud-based version of Magentic-UI, the company is Hosting It is on its Azure AI Foundry Labs.
