Manus markets a fully autonomous agent from China

Chinese start-up Butterfly Effect launched Manus, an invite-only “general AI agent” it billed as able to work autonomously on complex tasks — resume screening, stock analysis, website building — rather than simply chatting. A launch demo video drew more than a million views within twenty hours, and co-founder Yichao “Peak” Ji described the product as “potentially a glimpse into AGI.” Manus was built on existing large language models, reportedly including Anthropic’s Claude and fine-tuned versions of Alibaba’s Qwen, rather than a model of its own.

Access was tightly rationed from the start, officially because of limited server capacity. That scarcity created a resale market for invite codes, which reportedly changed hands for sums up to $13,800. Independent reaction split sharply: some early testers, including a Hugging Face staffer, called the results “mind-blowing,” while a TechCrunch reporter found it failed at basic tasks such as booking flights or ordering a sandwich. Security researchers also raised concerns about the confidentiality of tasks processed on Manus’s cloud infrastructure.

The launch mattered less for what Manus itself could do — reviewers could not agree on that — than for showing that the agent framing dominating US product launches that spring had a Chinese counterpart generating comparable hype, built largely by wrapping other labs’ models. Butterfly Effect later relocated much of its team to Singapore, and Meta’s subsequent attempt to acquire the company was blocked by Chinese regulators.