News
Microsoft Unveils Magnetic-One: A Multi-Agent AI System for Automating Complex Tasks
Microsoft researchers have introduced Magnetic-One, a new open-source, multi-agent AI system designed to automate tasks that typically require human intervention. The system is geared towards handling intricate, open-ended tasks across web and file-based environments, with the potential to enhance efficiency for enterprise users.
"It’s the difference between generative AI recommending dinner options to agentic assistants that can autonomously place your order and arrange delivery," wrote Microsoft researchers. "It’s the shift from summarizing research papers to actively searching for and organizing relevant studies in a comprehensive literature review."
Magnetic-One leverages a central orchestrator agent to coordinate four specialized agents: WebSurfer for internet navigation, FileSurfer for managing files, Coder for coding tasks, and ComputerTerminal for command-line operations. This orchestrated approach enables the system to perform complex, multi-step tasks autonomously, a leap in AI functionality, according to Microsoft.
It’s the evolution from AI merely suggesting options to autonomously executing multi-step actions, Microsoft researchers explained, highlighting the contrast between traditional AI recommendations and the agentic capabilities of Magnetic-One.
"The orchestrator plans, tracks progress, and re-plans to recover from errors, while directing specialized agents to perform tasks like operating a web browser, navigating local files, or writing and executing Python code," wrote researchers with Microsoft AI.
Though the system is still below human-level performance and susceptible to errors, Microsoft sees this release as a step toward more capable agentic AI. To address potential risks, the company is also releasing AutoGenBench, an evaluation tool designed to test the AI’s performance on complex tasks with controlled, repeatable benchmarks.
"Our criterion for selecting benchmarks is that they should involve complex multi-step tasks, with at least some steps requiring planning and tool use, including using web browsers to act on real or simulated webpages," the researchers wrote. "We consider three benchmarks in this work that satisfy this criterion: GAIA, AssistantBench, and WebArena."
Magnetic-One, available now for download, allows researchers and developers to experiment with and contribute to the evolving field of agentic AI, which Microsoft hopes to make safer and more reliable through community feedback and continuous testing.
Those looking to test drive Magnetic-One can download it here.