What is "Manus," the General-Purpose AI Agent That Shocked the World? | 2025/03/10, Vol.62

In April 2023, shortly after the release of GPT-4, an AI agent called Godmode also became a hot topic, but it often fell into an infinite loop just by breaking down tasks. The performance of the general-purpose AI agent "Manus" announced last week has reached a level unimaginable at that time. We will introduce the most impactful news and noteworthy tools announced recently!

1. Top 3 Recent Big News

① "I felt AGI for the first time" — Chinese startup releases general-purpose AI agent "Manus"

A Chinese startup has released the general-purpose AI agent "Manus", shocking the world. There are many voices saying, "I felt AGI for the first time," and it is so popular that invitation codes are being traded for hundreds of thousands of yen or more. Please take a look at the demo video below.

Manus is described as "an agent that combines Operator, Deep Research, and Claude Computer Use," and it can consistently execute web browsing, research, and programming.

The official website features actual use cases along with output examples. Since a wide range of operations is permitted, it is clear that it can handle an extremely wide range of real-world tasks.

Manus is currently provided as a closed beta version via invitation code, and it is highly rated by users, although it is mainly influencers. Rowan Cheung, who runs the AI media "Rundown AI," had Manus create his own self-introduction page, and after appropriately researching various SNS accounts and web articles, he rated it as "100% accurate content".

In the next section, we will consider the short-term and long-term impact that this Manus will bring.

② Possibilities of Manus — Parallel execution by AI agents will be a game changer

Information about Manus is still limited, but there is no doubt that its "high versatility" stands out among the agents that have appeared so far stands out. Cross-functional tasks that previously could not be realized without combining multiple tools and programs can now be executed just by giving instructions in natural language.

On the other hand, judging from the output examples posted on the official website and user reports, the main added value is the "visually easy-to-understand and organized" part, such as generating structured reports, visualizing data analysis results, and displaying them in HTML. There are few cases where it produces deep output based on its own thinking like DeepResearch, and it is unclear how much Manus can contribute to work efficiency. Also, it seems to take 2 hours to create comprehensive teaching materials, and slow execution speed has also been pointed out as a drawback.

In the medium to long term, what kind of impact will such general-purpose AI agents have? As OpenAI CEO Sam Altman previously stated regarding AI agents, "Think about AI not just making a restaurant reservation, but contacting 300 restaurants to find the best one for you,"as he put it,parallel execution by AI agents has the potential to be a game changer.

The video below claims to show 50 tasks being executed in parallel using Manus. You can probably imagine a future where general-purpose AI agents perform tasks at a speed that no amount of human labor could match.

According to Manus's privacy policy, the operating company is BUTTERFLY EFFECT, famous for developing "Monica," a Chrome extension that allows easy access to LLMs like GPT-4 in the browser. While there is little reliable information regarding the technology used, according to Sina Finance, which is operated by a major Chinese internet company, the firm does not develop its own LLMs and is merely an AI wrapper that integrates them.

While technical details about Manus will surely emerge in the future, we should also pay attention to how much of an advantage major AI companies like OpenAI, Anthropic, and Google still retain; this is a trend to keep a close eye on.

③ "Not quite past the uncanny valley, but approaching that threshold" — An extremely realistic AI voice has arrived

A startup founded by former Oculus CEO Brendan Iribe and others, "Sesame", has announced an AI voice companion capable of extremely natural conversation. Olive Moore, a partner at a16z who announced their investment, describes it as "more human than human". Please listen to an actual conversation via the link below.

From intonation, pacing, and tone to the way it breathes, it is as realistic as a human. One could say its level of perfection even surpasses OpenAI's "Advanced Voice Mode".

Sesame lists the following four elements as crucial for achieving realistic AI voice interaction.

In other words, only when an AI can understand the other party's emotions and situation, possess a "personality," and express itself at the right time, can a user truly feel they are "conversing with an AI." The company reports that it achieved this cutting-edge performance by training on approximately 1 million hours of audio data, primarily in English.

Even more interestingly, Sesame has set "breathing life into computers" as its ultimate goal, and is working not only on voice synthesis technology but also on the design and development of AI glasses. For them, voice synthesis technology is merely a means to an end, and they have announced that they plan to release the developed model as open source under the Apache 2.0 license.

While voice synthesis technology has made such progress, the current reality is that the quality of Japanese AI voice is several steps behind English. One reason for this is the unique structure of Japanese reading and accent determination. For example, there is the complexity where the accent differs between the place name "Shibuya" and the personal name "Shibuya".

Sesame has revealed plans to expand to over 20 supported languages within the next few months, but there are inevitably limits to "just using" overseas models as they are, and the development of domestic voice synthesis technology is eagerly awaited.

2. Picking up AI tools that are trending on social media!

Sesame (https://www.sesame.com/research/crossing_the_uncanny_valley_of_voice)

• A demo where you can converse with an extremely realistic AI voice that has been called "more human than human"

• Refined down to the smallest details, including intonation, pacing, tone, and breathing

• You can talk to either Maya (female) or Miles (male).

• Available for free trial. Note, however, that it does not support Japanese.

Hedra (https://www.hedra.com/)

• A Japanese-compatible lip-sync AI that can make characters in images speak in time with audio.

• The model has been updated to "Character-3", significantly improving quality. The awkward, jerky movements have been greatly reduced.

• Furthermore, emotions and movements can now be controlled via prompts.

• Even with the free plan, you can generate five videos of 30 seconds or less per day.

Introduction of the Operating Company

Mavericks Inc.Mavericks is a product-led AI startup. They possess numerous cutting-edge proprietary products, such as the video generation AI "NoLang" and the super-resolution technology "Kakudai".

In particular, NoLang is a video generation AI service that allows anyone to easily create high-quality videos from text or PDF documents, and it has surpassed 100,000 registered users. Videos are generated in an astonishingly short amount of time, so please give it a try.

Additionally, we deliver a wide range of AI news every week for free, covering everything from trend analysis of notable domestic and international companies to introductions of the latest AI. You can view past back issues below.