Manus, an invite-only ‘general AI agent’ developed by Chinese startup Butterfly Effect, has taken the internet by storm since its launch on March 5, 2025. The product promises to autonomously complete complex tasks such as building websites, generating slides, or booking flights with minimal human intervention. Demos on social media show Manus browsing the web, writing code, and producing polished outputs, leading some to call it a ‘second DeepSeek moment’ for Chinese AI. The hype has driven invite codes to sell for thousands of yuan on secondary markets, and the official Discord server grew to over 138,000 members in days.

However, a wave of skepticism has followed. Critics argue that Manus is essentially a polished wrapper around existing large language models like Claude and Qwen, orchestrated via tools and browser automation — not a new model developed in-house like DeepSeek. Early users report mixed results. TechCrunch tested Manus and found it crashed when trying to order a fried chicken sandwich, failed to book a flight to Japan (only serving broken links), and errored out while building a Naruto-inspired fighting game. AI researcher Alexander Doria encountered error messages and endless loops. Others note factual mistakes and inconsistent citations. The company acknowledges these issues, stating the closed beta aims to ‘stress-test various parts of the system.’

Despite the shortcomings, the hype persists. Influencers spread misleading demos, including one video showing a desktop program acting across smartphone apps — which the company confirmed was not actually Manus. While some see it as a practical step toward agentic AI, the question remains: is this a genuine leap or a well-timed hype cycle fueled by exclusivity and scarcity?