Since its launch last week, Manus, a general AI agent developed by the Wuhan-based startup Butterfly Effect, has gained rapid international attention, with tech leaders like Jack Dorsey and Victor Mustar praising its capabilities. Often compared to the groundbreaking DeepSeek, Manus claims to be the first general AI agent that operates autonomously across diverse tasks by utilizing multiple AI models and independent agents, setting it apart from traditional single-model AI chatbots. Currently, access to Manus remains limited, with under 1% of waitlisted users receiving invitation codes, although interest is evident with over 186, 000 members in its Discord channel. The MIT Technology Review managed to test Manus, finding its performance akin to that of a smart intern: adaptable and responsive but occasionally lacking in understanding and thoroughness. The design is user-friendly, with English as the default language, and users interact via a platform resembling ChatGPT. Testing Manus with three tasks, the results yielded mixed outcomes: 1. **Reporters List**: Manus initially provided a limited list of five journalists but quickly complied with feedback, offering a more comprehensive thirty-name list. Importantly, downloadable outputs make it easy for users to edit and share results. However, it faced challenges with paywalls during its research. 2.

**Property Search**: For an apartment hunt in NYC with specific criteria, Manus struggled initially but refined its output after clarification. The final list featured organized recommendations and took less time than creating the reporters list. 3. **Innovators Under 35 Nominations**: Manus faced the most complexity here, attempting to source 50 candidates for MIT Technology Review's list. It navigated numerous sources but was limited by paywalls and ended up producing a skewed list after several hours of work. In summary, while Manus demonstrated itself as a useful, potentially powerful tool—particularly for analytical research—it also exhibited instability, a higher failure rate compared to competitors, and struggled with large text inputs. While it’s still developing, especially in server infrastructure, its transparency and collaborative approach stand out. This tool is indicative of how Chinese AI firms are carving their own paths in the global AI landscape rather than merely imitating Western models. Overall, Manus shows significant promise for various practical applications in both professional and personal contexts.