Manus, an “agentic” AI platform that just launched in preview, is making waves—perhaps even more than a Taylor Swift concert.
The buzz around Manus is intense. The head of product at Hugging Face called it “the most impressive AI tool I’ve ever tried,” while AI policy researcher Dean Ball went even further, dubbing it “the most sophisticated computer using AI.” On Discord, the Manus community exploded, amassing over 138,000 members in mere days. Meanwhile, invite codes are reportedly selling for thousands on Xianyu, a Chinese resale platform.
But Manus isn’t built entirely from scratch. Reports suggest it runs on a mix of fine-tuned AI models, including Anthropic’s Claude and Alibaba’s Qwen. Users have leveraged it for tasks like drafting research reports and analyzing financial filings.
The company behind Manus, The Butterfly Effect, has made some bold claims on its website—suggesting the platform can do everything from purchasing real estate to programming video games.
In a viral video on X, Yichao “Peak” Ji, a research lead at Manus, positioned the platform as a superior alternative to agentic AI tools like OpenAI’s deep research and Operator. He claimed Manus outperformed deep research on GAIA, a benchmark evaluating an AI’s ability to navigate the web, use software, and execute complex tasks.
“[Manus] isn’t just another chatbot or workflow,” Ji said. “It’s a completely autonomous agent that bridges the gap between conception and execution […] We see it as the next paradigm of human-machine collaboration.”
But not everyone is convinced.
Alexander Doria, co-founder of AI startup Pleias, reported encountering error messages and infinite loops while testing Manus. Others on X noted its tendency to make factual mistakes, omit citations, and miss readily available information.
I tested it myself with what seemed like simple tasks. First, I asked Manus to order a fried chicken sandwich from a top-rated fast-food spot in my delivery range. After ten minutes, it crashed. A second attempt got me a relevant menu item, but Manus couldn’t complete the order—or even generate a checkout link.
Next, I requested a flight from NYC to Japan, specifying “business class, prioritizing price and flexible dates.” The best Manus could do? A handful of airline links, some of which were broken.
Hoping for a win, I tried something different: booking a table at a nearby restaurant. Manus failed again. Then, I challenged it to build a Naruto-inspired fighting game. Half an hour in, it errored out. At that point, I gave up.
A spokesperson for Manus responded to TechCrunch via DM:
“As a small team, our focus is to keep improving Manus and make AI agents that actually help users solve problems […] The primary goal of the current closed beta is to stress-test various parts of the system and identify issues. We deeply appreciate the valuable insights shared by everyone.”
So if Manus is struggling to meet expectations, why is it generating so much hype? A mix of factors played a role.
The platform’s invite-only model created an aura of exclusivity. Chinese media quickly hailed it as a breakthrough, with QQ News calling it “the pride of domestic products.” Meanwhile, AI influencers amplified misinformation about Manus’ capabilities. A widely shared video supposedly showed Manus operating multiple smartphone apps autonomously—but Ji later confirmed that wasn’t a real demo.
Some AI enthusiasts also likened Manus to DeepSeek, a Chinese AI company known for its in-house models and open-source contributions. But the comparison is shaky—unlike DeepSeek, The Butterfly Effect hasn’t built proprietary models or released its tech to the public.
To be fair, Manus is still in its early days. The team says they’re actively scaling computing power and fixing reported issues. But as it stands, the platform seems to be a classic case of hype outpacing reality.
