Have you heard of Manus, the AI agent being hailed as the next DeepSeek and taking the world by surprise?
Launched in early March 2025, Manus is a general-purpose AI agent developed by Butterfly Effect, a startup based in Wuhan, China.
It drew even more attention after receiving endorsements from notable tech figures, including Twitter co-founder Jack Dorsey and Victor Mustar, product lead at Hugging Face.
So what exactly makes Manus different from the AI models that came before—and why is it earning such high praise?
A general AI agent refers to an artificial intelligence system that isn’t limited to a specific task or function, but instead can understand and carry out a wide range of activities on its own.
While traditional AI systems are typically designed to answer questions or perform one specialized function, a general AI agent can interpret user intent, retrieve relevant information independently, and complete multi-step tasks. Much like a human planning and executing complex work, it can make decisions and handle various assignments flexibly.
For example, beyond simply providing information, a general AI agent could plan a travel itinerary, generate a phrasebook tailored to that trip, or search for and recommend real estate listings—all autonomously.
Ultimately, general AI agents serve as digital collaborators, taking on repetitive and complex work while assisting humans in more dynamic, interactive ways.
Most AI services we’re familiar with—like ChatGPT or DeepSeek—operate on a single large language model (LLM). They typically function through a conversational interface, providing information or responding to user queries.
Manus is different.
Rather than relying on a single model, Manus operates autonomously by coordinating multiple AI models and independent agents that work together to complete tasks. For example:
Claude 3.5 Sonnet from Anthropic
A fine-tuned version of Qwen, the open-source model released by Alibaba
These diverse models collaborate under the hood, allowing Manus to interpret user instructions, plan accordingly, and carry out complex, multi-step tasks on its own. It’s not just a chatbot—it’s more like an autonomous digital intern or assistant.
1. It breaks down and completes complex tasks on its own
Manus can autonomously divide complex tasks into manageable steps and execute them independently. Whether it’s compiling a list of journalists, searching for real estate listings, or generating talent recommendations, it navigates the web, gathers information, analyzes it, and delivers results—all without being explicitly told how to do each step.
2. It shows you the entire workflow in real time
One of Manus’s standout features is the “Manus’s Computer” window. Displayed on the right side of the screen, this panel lets users watch Manus work in real time—step by step. If needed, users can even step in mid-task to adjust or intervene. It’s like sitting next to a digital assistant and giving it live feedback as it works.
Manus carrying out a user command, with each step visible through the "Manus’s Computer" panel on the right. Source: Manus
3. Fast at incorporating feedback—and great at learning
Manus learns from user feedback and applies it to similar tasks in the future. Once you give it guidance, it remembers and continuously improves. You can also replay or share each session.
Of course, it’s not perfect yet. According to hands-on testing by the MIT Technology Review team, there are still a few issues:
Frequent server overloads
Users often see the message: “The service is currently overloaded. Please try again later.”Performance drops with large tasks or long input texts
In some cases, Manus slows down or stalls when handling heavier workloads.User intervention needed for web access
CAPTCHA challenges or paywalled content may require the user to step in manually.
Still, if server reliability improves, Manus could become an incredibly valuable tool—especially for freelancers, startups, and small teams.
Manus isn’t just a chatbot—it’s a full-fledged AI agent.
AI agents have already established themselves as smart collaborators that boost productivity by handling repetitive and time-consuming tasks. At SelectStar, we also offer solutions powered by AI agents.
But for AI to be truly useful, the outputs it generates must be accurate, trustworthy, and free from harmful biases or misinformation. That’s why SelectStar developed Datumo Eval, an AI agent–powered solution for evaluating the safety and reliability of LLMs.
Datumo Eval
👉 A solution for evaluating the quality and safety of responses generated by large language models (LLMs)
Datumo Eval helps assess and improve whether AI-generated responses are factually accurate, easy to understand from a user’s perspective, and free from harmful bias or misinformation. It automatically generates question datasets and uses custom-defined metrics to evaluate responses based on your specific evaluation goals.
Multiple AI agents are assigned to different roles in this process, collaboratively producing sharp, targeted questions that drive more rigorous assessments.
Datumo Eval is a must-have tool for proactively auditing these risks and ensuring the trustworthiness and quality of your AI services.
