“Innovation or Hype?” China’s Startup Unveils Autonomous AI ‘Manus AI’

Photo of author

By Global Team

The AI agent ‘Manus AI’, developed by the Chinese startup Butterfly Effect, is gaining attention as a new turning point in autonomous AI technology. According to a report by TechWire Asia, this AI is introduced as the world’s first general-purpose AI agent capable of handling complex, multi-step tasks with minimal human intervention.

Manus AI, based on a multi-agent system, processes complex workflows through collaboration among various sub-AIs and is integrated with external tools such as web browsers, code editors, and databases. Its multimodal capabilities that handle diverse data like text, images, and code, and its adaptive algorithm that learns progressively based on user feedback, are highlighted as key features.

Co-founder of Manus AI (Butterfly Effect), Jiichao, introduces the company's AI agent 'Manus' released on the 5th.
Co-founder of Manus AI (Butterfly Effect), Jiichao, introduces the company’s AI agent ‘Manus’ released on the 5th.

Performance-wise, it is also generating anticipation. In the GAIA (Generalized AI Agent) benchmark, which evaluates the actual problem-solving ability of AI agents, Manus AI outperformed OpenAI’s Deep Research system, scoring 86.5%, 70.1%, and 57.7% in basic, intermediate, and complex tasks respectively, compared to OpenAI’s 74.3%, 69.1%, and 47.6%. However, it showed a pattern of performance dropping significantly as tasks became more complex, indicating difficulties even advanced AI models face with sophisticated multi-step reasoning.

Performance evaluation results of Manus AI in the GAIA benchmark.
Performance evaluation results of Manus AI in the GAIA benchmark.

The user interface emphasizes transparency and control of AI. Through the ‘Manus Computer’ window, users can observe the AI’s decision-making process in real-time and can intervene to adjust AI actions at any time during the task. It also introduces asynchronous task processing and an intuitive chat interface, aiming to make collaboration with AI more natural.

However, initial tests consistently raised performance issues. Even with simple tasks such as ordering food or booking flights, system crashes and infinite loops were reported, and some users pointed out a lack of reliability in results. Particularly, as Manus AI is built on existing AI models from Anthropic and Alibaba, there are doubts about whether there is any unique technological innovation.

Critics suggest that the rise of Manus AI might be driven more by marketing and exclusivity strategies than by genuine technical innovation, asserting that it will take time to prove its true practicality. As the AI agent market grows rapidly, there is keen interest in whether Manus AI can demonstrate real innovation through continuous development.

Leave a Comment