Manus, an "autonomous" AI platform previewed last week, has generated more buzz than Taylor Swift's concerts.
The product lead at Hugging Face described Manus as "the most impressive AI tool I've ever used." AI policy researcher Dean Ball called it "the most sophisticated AI computer." Within days, the official Discord server for Manus grew to over 138,000 members, with invitation codes reportedly being resold on China's Xianyu app for thousands of dollars.
However, it remains unclear if this hype is justified.
Manus wasn't built entirely from scratch. According to social media reports, the platform uses a combination of existing and fine-tuned AI models, including Anthropic's Claude and Alibaba's Qwen, to perform tasks such as writing research papers and analyzing financial documents.
Nevertheless, Monica, the Chinese startup behind Manus, showcases some remarkable examples of what the platform allegedly can do on its website, ranging from purchasing real estate to programming video games.
On X, Yichao "Peak" Ji, the research lead of Manus, hinted that the platform outperforms OpenAI's Deep Research and other autonomous tools like Operator. Ji claimed that Manus surpasses Deep Research in GAIA, a popular benchmark test for general AI assistants that evaluates an AI’s ability to execute tasks by browsing the web and using software.
"[Manus] is not just another chatbot or workflow," Ji stated in a video. "It's a fully autonomous agent bridging the gap between ideation and execution [...] We see it as the next paradigm of human-computer collaboration."
But some early users have pointed out that Manus isn't flawless.
Alexander Doria, co-founder of AI startup Pleias, mentioned encountering error messages and endless loops while testing Manus in a post on X. Other X users noted that Manus often made factual mistakes, inconsistently cited sources, and frequently missed information readily available online.
My own experience with Manus wasn't particularly positive.
I asked the platform to handle what seemed like a fairly simple request: ordering a fried chicken sandwich from top fast-food restaurants within my delivery area. About ten minutes later, Manus crashed. On the second attempt, it found menu items matching my criteria but couldn't complete the ordering process—not even providing a checkout link.
When I tasked Manus with booking a flight from New York to Japan, it failed again. Despite giving instructions I thought were straightforward (e.g., "look for business-class flights prioritizing price and flexible dates"), all Manus could do was provide links to airline websites and flight search engines like Kayak, some of which were invalid.
Hoping for better results, I instructed Manus to book a table for one at a nearby restaurant. It failed after several minutes. Then I asked the platform to create a Naruto-style fighting game. After half an hour, it encountered an error, prompting me to give up.
A spokesperson for Manus sent the following statement to TechCrunch via DM:
"As a small team, our focus is on continuously improving Manus and creating an AI agent that truly helps users solve problems [...] The primary goal of the current closed beta is to stress-test various parts of the system and identify issues. We greatly appreciate the valuable insights shared by everyone."
So why has Manus gone viral if it fails to meet its technical promises? Several factors contributed to this phenomenon, such as the exclusivity created by limited invitations.
To be fair, Monica's Manus is still in the early access phase. The company claims it's working hard to expand computing power and fix reported issues. However, based on the platform's current state, it seems that the hype surrounding Manus has outpaced technological innovation.
