Manus's victory is a victory for the product

Manus Latin, hand.

Manus.im is the most popular general agent product this week.

Unlike other large language models, Manus not only uses its brain and mouth, but also its hands . It can not only chat with you , but also call various interfaces to deliver results to you.

The gust of wind has its origin. Before talking about Manus, we can look back at the three AI  products in the programming industry in 2024, namely: Cursor, Windsurf, and Devin.

Cursor: A programming CoPilot tool for programmers. With the support of AI, it can help programmers complete code writing, completion and prediction faster.

WindSurf: It is closer to a programming agent, which drives the program through dialogue. It does not rely on your programming knowledge. As long as you put forward your requirements in the dialog box, it can program itself, and it has opened up the terminal capabilities, can debug itself, and install various missing libraries.

Devin: An independent agent. Devin first built the planning capability (Planner ), and then it has its own independent editor and browser, which can retrieve information and program by itself.

When using Cursor, you need to know coding, which gives you more control; when using Windsurf, you don’t need  to know coding, but you still have to install VSCode (programming environment); when using Devin, you don’t need to know anything or install anything, you just need to spend money.

So, some people joked: If you are a programmer, you will choose Cursor; if you are a newbie, you will choose WindSurf; if you are a boss, then try Devin as soon as possible.

When we compare these three products, it is not difficult to find that from the delivery process to the delivery results, it is an inevitable process for a product to move from a small vertical group to the general public.

After talking about these three products, let’s talk about the reasoning chain of DeepSeek and OpenAI O1.

The most touching thing about DeepSeek from a product perspective is not just its effect, but also that it fully displays the thinking process. In comparison, OpenAI released o3's processed thinking chain only after DeepSeek, which caused a lot of criticism.

From process to result, from black box to transparent, Manus has both product features, no wonder it has set off a wave. Judging from the cases on Manus's official website, many projects can be done without Manus, and I have built some of these projects myself using Coze's workflow in series.

However, the highlight is to let AI act as its own planner, run remotely, and perform multiple functions in series, delivering results rather than improving process efficiency.

It allows more users without any basic knowledge to experience AI Agent without any barriers: you throw the task to AI, it performs the task remotely, and notifies you for acceptance after completion.

If you want, you can also watch the whole process of Manus's work (as shown below). It will conscientiously split multiple TODO items, tick the completed items one by one, open web pages to browse information, and call interfaces to draw pictures. The whole process can also be replayed, and the completeness of this work daily report is unparalleled.

Many people question whether Manus is a shell or a patchwork.

You are all right, so what?

C-end users don’t care about what you said.

I subjectively infer that in the process of developing Monica, the company has accumulated a large amount of user preference data and connected to a large number of tools and interfaces (drawing, programming, file deployment, and browser capabilities). It is the accumulation of these data and the precipitation of hard work that allow Manus to better understand users and seamlessly dispatch more interfaces, thereby providing a smooth end-to-end experience.

Because Manus requires a queue, I tried OpenManus in parallel. It was so bumpy that it drove me crazy. It repeatedly failed to call the browser capabilities. It may be a good technical framework, but it is definitely not a qualified user product.

Technology is important, but not that important. Faster iterations, better product experience, and allowing more users to use it are what are truly most important.

There are many parts that can be iterated in the process of applying Manus:

For example, for user account authorization, I can mount my account authorization to Manus, so that it can bypass the registration wall and access more data content.

For example, it allows adding CheckPoints, so that users can take a quick look at the process before continuing with complex tasks , rather than running for several hours at a time.

For example, obtaining information from parsing web pages to reading screenshots, just like Musk's insistence on visual solutions rather than radar solutions, makes Manus more adaptable to this world constructed for humans.

But isn’t AI an evolving product?

Manus's victory is a victory for the product:

The paradigm of AI has changed from "what process do I require AI to perform" to "what results do I require AI to deliver".