The “agentic” AI platform Manus, which debuted in preview last week, is creating more buzz than a Taylor Swift performance.
Hugging Face’s head of product referred to Manus as “the most impressive AI tool I’ve ever tried.” The “most sophisticated computer using AI” is Manus, according to AI policy scholar Dean Ball. In a matter of days, the official Manus Discord server reached over 138,000 members, and invite codes for the group are apparently fetching thousands of dollars on the Chinese reseller app Xianyu.
However, it’s unclear if the excitement is warranted.
excellent
— jack (@jack) March 9, 2025
Manus wasn’t created from the ground up. Social media sources claim that the platform combines pre-existing and refined AI models, such as Alibaba’s Qwen and Anthropic’s Claude, to carry out duties including creating research reports and examining financial files.
However, the Chinese business that created Manus, The Butterfly Effect, provides some outrageous examples of what the platform is allegedly capable of on its website, like purchasing real estate and creating video games.
Yichao “Peak” Ji, a research lead for Manus, suggested in a widely shared video on X that the platform was better than agentic tools like Operator and OpenAI’s deep research. Ji asserted that Manus surpasses extensive study on GAIA, a widely used benchmark for general AI assistants that examines an AI’s capacity which probes an AI’s ability to carry out work by browsing the web, using software, and more.
Ji stated in the video that “[Manus] isn’t just another chatbot or workflow.” It is an entirely independent agent that connects the dots between ideation and implementation […]. We believe it to be the next big thing in human-machine cooperation.
However, some early adopters claim that Manus is not a cure-all.
In a post on X, Alexander Doria, co-founder of Pleias, an AI business, claimed that when testing Manus, he ran into infinite loops and error warnings. Other X users noted that Manus frequently overlooks material that can be readily obtained online, makes errors on factual queries, and fails to properly credit its work.
Deep Research took less than fifteen minutes to complete. Regretfully, Manus AI failed at step 18/20 after 50 minutes! 😑 I was observing Manus’ output, and it appeared to be doing very well. However, because it takes too long, it can be a little annoying to execute the same prompt twice!
— Derya Unutmaz, MD (@DeryaTR_) March 8, 2025
Personally, I haven’t had a great experience with Manus.
Ordering a fried chicken sandwich from a highly regarded fast food restaurant within my delivery radius sounded like a very simple request, so I requested the platform to handle it. Ten minutes or so later, Manus crashed. It discovered a menu item that satisfied my requirements on the second try, but Manus was unable to finish the ordering procedure or even supply a link for the checkout.
When I requested Manus to arrange a flight from New York City to Japan, it also whiffed. The best Manus could do, given instructions that, in my opinion, left little opportunity for interpretation (e.g., “look for a business-class flight, prioritizing price and flexible dates”), was to present links to fares across a number of airline websites and airfare search engines, some of which were broken.
I instructed Manus to book a table for one at a nearby restaurant, hoping that the next responsibilities would be the deciding factor. After a few minutes, it failed. I then requested that the platform create a fighting game using Naruto themes. I decided to give up after it failed to function for thirty minutes.
Honest opinion after trying Manus AI for the last 3 days, here’s the good and the bad.
Good:
– The research it does on the internet and the reports it generates are incredible.
– Its ability to run scripts behind the scenes to execute tasks is impressive.
– The plans it…— AshutoshShrivastava (@ai_for_success) March 9, 2025
TechCrunch received the following statement via direct message from a Manus representative:
Our goal as a small company is to continuously enhance Manus and create AI agents who genuinely assist people in resolving issues […]. The current closed beta’s main objective is to find problems by stress-testing different system components. We are incredibly grateful for everyone’s insightful contributions.
So why did Manusis blow up if it isn’t living up to its technical promises? A limited number of invites produced exclusivity, which was one of the contributing causes.
Manus was quickly hailed as an AI innovation by Chinese media, with magazine QQ News referring to it as “the pride of domestic products.” In the meantime, false information on Manus’ capabilities was disseminated by AI influencers on social media. In a widely circulated video, a desktop application—possibly Manus—was seen acting across several smartphone apps. Ji affirmed that the video wasn’t actually a Manus demo.
Other well-known AI accounts on X attempted to make unfounded analogies between Manus and the Chinese AI startup DeepSeek. Unlike DeepSeek, The Butterfly Effect did not create any internal models. Furthermore, although DeepSeek made a lot of its technologies publicly accessible, Monica hasn’t, at least not yet.
Manus is in very early access, to be fair to The Butterfly Effect. The business asserts that it is attempting to increase processing power and address problems as they arise. However, as the platform stands right now, Manus seems to be an example of hype outpacing technological advancement.