Microsoft has introduced Magentic-One, an open-source, multi-agent AI framework designed to autonomously manage complex digital tasks. Unveiled at Build 2025, the system aims to enhance productivity by coordinating specialised agents to perform intricate, multi-step operations across various domains.
At the heart of Magentic-One is the Orchestrator, a central agent that supervises four specialised sub-agents: WebSurfer, FileSurfer, Coder, and ComputerTerminal. WebSurfer navigates and interacts with web browsers; FileSurfer manages file-related operations; Coder writes and analyses code; and ComputerTerminal executes code and performs system-level tasks. This modular architecture allows each agent to function independently while collaborating to complete complex workflows.
Built on Microsoft’s open-source AutoGen framework, Magentic-One is model-agnostic, supporting various large language models , including GPT-4o. This flexibility enables developers to tailor the system to specific needs, balancing performance and cost. Microsoft has also released AutoGenBench, a benchmarking tool to evaluate the performance of agentic systems like Magentic-One against tasks requiring multi-step planning and tool use.
In performance evaluations, Magentic-One demonstrated competitive results on benchmarks such as GAIA, AssistantBench, and WebArena, showcasing its capability in handling tasks that demand inter-agent coordination and adaptability. These benchmarks assess the system’s proficiency in managing complex, real-world scenarios, indicating its potential applicability in various industries.
Microsoft has emphasised the importance of safety and responsible deployment of Magentic-One. The company recommends running agents in isolated containers to prevent direct attacks and monitoring logs to detect and mitigate risky behaviours. Human oversight is advised, especially for high-stakes actions, and access to sensitive data should be restricted. These precautions align with Microsoft’s commitment to responsible AI principles, ensuring that the deployment of Magentic-One adheres to ethical standards.
The introduction of Magentic-One positions Microsoft alongside other tech giants exploring multi-agent AI systems. Competitors like Salesforce’s Agentforce and IBM’s Bee Agent Framework are also developing solutions to automate complex tasks traditionally handled by humans. Microsoft’s approach, focusing on modularity, flexibility, and safety, aims to provide a robust platform for enterprises seeking to integrate AI into their operations.