Manus's explosive red and black: the marketing is indeed excessive, but the product is not fake

To be honest, there is no invitation code for manus until now, so if you insist that you don't have a say if you don't experience it, just fork it out can save you ten minutes.

I want to talk about two big points, one is about the marketing controversy of Manus; One is Manus's approach to product ingenuity.

Product: There was no breakthrough, but there was a breakthrough

Manus has not achieved a breakthrough in technology, which is perhaps the biggest consensus after the controversy. The core verification case comes from the MetaGPT team (the programming agent team) who copied an OpenManus in three hours.

But there is no doubt that the products that Manus brings to you are shocking. It uses AI as a "human hand" to automate a series of processes, including a large amount of information self-collection, browser interaction, and so on. These will eventually be encapsulated, and the user will simply tell Manus what he wants to do, and then move a chair and watch the show directly.

If you're like me and follow some of the latest developments in the AI space, you'll find that manus is more of a stitch monster than a deepseek model training innovation:

1. Task sorting and knowledge base calling. Most models have this capability, and from the perspective of prompt engineering, disassembling and sorting out the task first will improve the effect of the final output of the AI. In my own day-to-day interactions with AI, I either make a list or let the AI sort it out for me based on my needs.

2. Information collection, sorting and analysis. That is, Deep Research, which is basically supported now.

3. External tool function call. Both MCP and a large number of open source tools such as browser-use have been incorporated into everyday use cases.

4. Multi-agent cooperation. It should have been a year or two since Devin started (I can't remember exactly when). What the metaGPT team that did the three-hour reproduction was a multi-agent collaboration in the field of programming.

A lot of people hold on to Manus' lack of innovation, and they start to sneer when they see the praise.

It is important to know that arrogance is the worst enemy of progress.

You might as well ask yourself: since they are all existing things, why do they explode all of a sudden when they are integrated? It's all something that's already there, why don't you sew it?

Everyone looks at innovation from a very narrow perspective, always grasping the technology, but they don't know that the innovation of product concepts and business models is more far-reaching and influential.

I put an article at the end, and the understanding of the Manus product is very well written (because it is paraphrasing the words of the team itself):

! Manus's Red and Black: The marketing is indeed excessive, but the product is amazing

This passage is also the main reason why I wrote this article. I think the points mentioned here are worth savoring for everyone who makes products.

Let's talk about what I've seen from Manus and what we can learn from it:

1. Anti-interference of agent workflow

Making games will also mention a similar concept, "flow".

That is, the user's experience of using the product is composed of a period of behavioral processes. For example, if I use Alipay to pay the phone bill, I need to open Alipay, click "Recharge" on the homepage, select the mobile phone number, select the amount, confirm, pay, and receive feedback on the successful recharge. In such a process, if there is an interference from an extra-process factor, the overall process will be interrupted and the product experience will decline.

Existing AI agents, while not completely exclusive to the user's workflow, do have the problem of taking up space in the workspace. Take the commonly used AI reading web page as an example, the AI gives an understanding based on the current web page, and you can't change the current web page, otherwise it will be invalid. Otherwise, copy the link manually and give it to the AI in a separate window. Another product I recently tried same.dev, which focuses on directly copying the front-end source code, has the problem of bypassing the web page and interrupting the operation, occupying a workspace alive, and the experience is very poor.

Of course, this is not to say which is better or worse. However, the changes at the product level of manus are meaningful for us to think about the evolution of the AI agent form.

From in-browser AI to AI-interactive page in-browser, the former satisfies the user's day-to-day needs (AI is only part of the work-life process), and the latter satisfies the needs of intelligent agents (reducing interference from extraneous factors).

The AI Agent product only needs to be an interactive page that receives input, displays the process, and spits out output. The process is only allowed to be displayed, and the user will not be disturbed by the process.

I'm reminded of what I mentioned in a previous article about the problem with the embedded browser of OKX Wallet. You can't let the user actively interrupt the process to achieve the goal.

The experience of an interruption of "flow" is terrible.

2. Rethink "External Tools"

In most people's eyes, only standard protocols such as MCP or packaged libraries are considered "add-ins".

In fact, the so-called tool is a black box with stable input and output, which allows users to have clear expectations.

Programming is the most deterministic one in AI application scenarios, and countless scripts and modules can be packaged as tools.

From the perspective of model training, it is no problem to compete and solve math problems; But if you take the math problem directly from the perspective of model application, it is stupid.

Why don't you write code to solve it, and you have to use a Rausch subvector map?

manus actually points out the key point in the design of general-purpose agents: don't try to solve all problems directly with AI.

AI is just hands.

Hand use tools to solve problems. Tools can be defined in advance or written on an ad hoc basis.

People - Hands - Tools - Tasks.

How many layers of indirect tools do you care?

3. Non-destructive to use scenarios

Students who often program must have a deep understanding that different projects and dependent library versions are messy, and as long as one of the key library versions is not correct, the project may run with errors.

This is where the need for a virtual environment in programming comes in. npm is based on project dependency installation, python creates virtual environments, docker containers, etc., all of which in my understanding are to ensure the independent customization of the environment.

This may be the consensus of all product-level agents with direct toC at this stage: don't encroach on the user's on-premises environment, use the cloud.

There were bolt.new, mgx.dev, etc., all of which chose to run directly in the cloud, write and debug. However, these are programmatic agents, and general-purpose agents lack controls.

In stark contrast to Manus's approach, there's another product, Highlight. After you download and install it, a floating window will be formed on the desktop, and there will be some AI integration operations based on the application of the current workspace.

Seems to be eye-catching?

For example, I don't know how to crawl, I go to the browser page and let Highlight crawl for me?

Personally, I've experienced that the practice of highlight has interfered with the original workflow because I have to open the process for highlight to operate. But in reality, anyone will switch pages back and forth at work, and it's impossible to wait specifically for the AI to complete. Also, if the AI is crawling with my browser, is it using my IP, and will it affect my future visits?

The local scene is ruined.

In fact, these points are not difficult to think of just by talking about them. But how to carry out a series of designs in a framework, I think it is still worth experiencing.

Finally, my personal expectation regarding the manus cap.

My expectation is to control the expectation – AI can't do everything for us. Even if AI modules are introduced into smart homes in the future, they will become "tools" that can be used by "human hands"; Even if there are more and more "tools" on the desktop that allow us to control production-grade software with only natural language; In many cases, humans also have to do the process control test, because the AI's perception of the world is built on a huge black box, and they will "hallucinate".

In any case, since we believe that the truth of humans is greater than the truth of AI, humans need to test the production of AI. And the more you give to AI, the more you have to examine. This eventually balances above a boundary.

This boundary is the upper limit of general-purpose agent products.

Marketing: Not afraid of controversy, afraid of no controversy

Looking at this wave of manus out of the circle, it is not so much a spontaneous controversy in the circle, but a pair of "hands of God" secretly guiding.

These hands of God must also be official. Wouldn't it be wonderful to lead the hand with one's hand?

First of all, I asked Grok to summarize the marketing events of manus during this period:

! Manus's Red and Black: The marketing is indeed excessive, but the product is amazing

It can be clearly seen that the official core slogan is "the world's first general-purpose AI Agent".

This is a very controversial statement.

A. For outsiders, this statement is very eye-catching;

B. For people like me who pay attention to daily life but are not insiders, it can be seen at a glance that they are playing a word trick: I have introduced the product level above, and it is a stitching monster, and it cannot be called the first in any case; But if you add the word "daily", there is indeed no such consumer product that says that it has to deal with general-purpose tasks, causing a lot of public opinion. In addition, the slogan is inherently exaggerated, so it is indeed understandable to say so;

C. As for the insiders, I think most of them may be more indignant, after all, the results of their own research have been sewn in, or that this thing in their eyes has no technical content, but he has stolen the limelight.

• Opposing positions have emerged.

Where there are opposing positions, there is controversy, and public opinion eventually continues to ferment, bringing a leverage effect to Manus's propaganda.

Don't look at appearances, look at results.

As a result, Manus has gained global attention. This marketing is full of cost performance.

and the invitation code mechanism.

While the attention is full, the invitation code is strictly limited, on the one hand, it is based on cost considerations, and on the other hand, it is also to cover up the lack of products. After all, the "general-purpose" boasted of Haikou, and once it was opened, it could be directly blown up by various bugs and feedback from Class A people. In that case, it will not be an opposing position, but a one-sidedness, and it is likely to be cold. From this point of view, the invitation code mechanism is actually similar to an early test, where as many dev bugs are fixed, as many codes are placed, and seed users can help improve the product first.

Then there's the hunger marketing plan. Hunger marketing is essentially a competition for attention, and it is normal for all those who use the invitation code to be scolded by those who "love but can't".

What happened afterwards, X account freeze, technology jailbreak, open source, etc., were regarded by many as the evil consequences of excessive marketing.

My personal opinion on this point is that the AI circle is still too literary. It is recommended to bring a walk in the crypto circle and learn some shameless spirit.

The manus team is actually okay, after all, the last product, Monica, has been stable and profitable, but many small teams are still struggling on the edge of the food and clothing line. At this time, you told me not to overdo marketing? If it is possible to achieve high marketing results at a low cost, I would like to ask why not?

Can the face support the R&D team to continue to do it? Can faces provide enough money for innovation?

Faces are worthless. In this age of entertainment to death and information explosion, attention is what counts.

The most cruel and tempering thing about the crypto circle is that it is so close to the money itself that all you can see is the most real human nature and the most bloody routines (not bloody routines, but you bloody routines).

You have to be principled, but you can't help but understand and learn to accept some non-principled realities.

Otherwise, you will die a miserable death.

Resources

Complete review: How was Manus born? | Geek Park

Behind the explosion of Manus, how can Agentic AI products build a lasting competitive advantage?