For domestic product managers, it's almost impossible to use moltbot.

Currently, when working as an AI product manager, a lot of new AI technology information first emerges in Silicon Valley and then spreads to China.

The recently viral Moltbot also followed the same pattern, starting from Silicon Valley and then reaching China. The entire spread process took nearly half a month.

Even though this product is so popular, it is of little use to product managers in China.

Molbot is more useful for overseas product managers: it offers open permissions for social information.

During an interview live stream, Molbot's founder mentioned that initially, he thought large - scale AI model companies would develop this kind of product. However, after waiting for half a year and seeing no one take action, he decided to do it himself.

The reason is actually quite simple. In daily use of apps like Telegram and WhatsApp, users cannot directly converse with AI models. Instead, they have to separately open general - purpose models, which is very inconvenient.

If the social software used daily could directly send messages through an AI model to control various personal applications, life would be much more convenient.

So, based on his own demand scenarios for AI models, he developed this product and open - sourced it.

Looking at the founder's background, even though he has a history of open - source projects and is extremely fond of open - source sharing of code, after taking a three - year break, he re - entered the AI field.

It's hardly useful for product managers in China

Currently, mainstream social tools like WeChat and QQ are closed systems. They do not support any APIs and do not allow third - party plugins to access chat data. Currently, only the API robots provided by enterprise WeChat and Feishu can obtain messages and relevant work - related information data.

So, deploying Clawbot on a company's Feishu or enterprise WeChat naturally serves as a "Jarvis" for employees within the enterprise rather than for individuals.

Data security and personal permissions

To become a personal assistant, for example, sending WeChat chat messages to Moltbot for control, currently, the data still needs to be uploaded to a public cloud or a company's official account or mini - program to be processed.

For automated office scenarios in China to work, there must be software like WPS, online documents from Tencent and Alibaba, and online design tools. However, these companies do not open their corresponding core data interfaces, which is incomparable to Notion.

In the form of online collaborative office, unlike local clients, Moltbot not only needs permission to access browser operations but also requires corresponding account passwords and human - machine verification.

This makes it difficult for product managers in China to find a reason to use Moltbot.

Commercialization in China is always the fastest, but what about the efficiency - to - cost ratio rather than just revenue? It's a big question mark.

After the open - source project was launched, Huaqiangbei has already launched hardware devices. Users can directly purchase the hardware to use Moltbot. However, due to the permission issues of domestic social products mentioned above, it is difficult to use, and it may even be a bubble.

What is truly effective for product managers is to be able to operate home digital systems and relevant IoT devices by sending WeChat messages. However, this set of functions has not been well - compatible with the domestic market yet.

Deploying on Macmini is not that simple

Even though there are many online tutorials and one - click foolproof deployment options in 30 minutes, Macmini is still not a good long - term paid option. The AI models deployed on Macmini are always inferior to the currently paid ChatGPT or even the full - fledged DEEPSEEK. After using it for a long time, you'll find it rather stupid.

For example, the recently open - sourced Kimi 2.5 is a model with trillions of parameters and requires two 512GB Mac Studios to run. Using just a Macmini will make your personal "Jarvis" extremely stupid, making you reluctant to use it.

Whether it's logical reasoning ability or handling long texts, it can't compare with the mainstream models you are currently using. Of course, you can connect to the API of paid AI models.

So, you can see that Moltbot can only serve as an efficiency management tool within enterprises. However, there is no clear revenue scale, which is why it's hyped up but product managers in China rarely use it in their daily work.

Some people use two Mac Studio Ultra 512GB to run Moltbot with Kimi 2.5

Overseas, some bloggers are trying to use two Mac Studio Ultra computers to run Moltbot with Kimi 2.5 to achieve the best intelligence for their personal agents. However, with an investment of nearly 100,000 RMB, how much efficiency can it really bring? If it's just for automatically sending messages on Telegram and WhatsApp, the cost - to - efficiency ratio is too low.

Of course, some bloggers have deployed it for personal health management and other task management. They also need to recharge and use paid services like Claude and ChatGPT API for invocation.

In short, currently, it's still unclear how much automated operation can achieve effective value, especially for product managers in China.

The AI robot forum: self - indulgence of the founding team

Currently, the AI robot forum seems extremely popular, with discussions that only others can watch. However, it turns out that it's just the agents built by the Moltbot founding team running on it.

In essence, current large - scale models are mainly for text understanding, and their understanding of space and the physical world is very limited, such as weight, pressure, wind speed, temperature, etc.

These require multi - modal generation of systematic models, which is still a long way off. Of course, for product managers in digital office, they are more than sufficient for generating PPTs, Word documents, and UI design drawings.

So, we need to demystify. Although Moltbot was extremely popular at the beginning of the year, we should understand that it represents a direction for AI product development, namely open - source, localization, and personalization. As for whether to invest in it now, the interest among product managers is currently almost zero.

For example, I'm currently arranging my team to deploy Moltbot, but it's only for team use, not for myself. Since I mainly use WeChat in China, using Moltbot with my WeChat account could get it blocked at any time. Why take the risk?

Moltbot and Kimi 2.5 are the dividing lines for the next product iteration of AI product managers

In the future, open - source will become the mainstream for AI models and a growth channel that AI product managers need to consider. Through open - source, after acquiring users, they can obtain paid B - end users. Various industries can deploy the models in their own data centers and computer rooms, thereby improving overall industry efficiency.

Rather than having data guided by others for training, going open - source allows for absolute control over data and model training capabilities.

Finally, why are good AI product managers extremely scarce?

I saw a post on Zhihu: "Why are good product managers extremely scarce?"

I think the main reason is that many product managers can't keep up with the times and are reluctant to actively understand the latest mainstream technologies and terminal devices. For example, now is a period of rapid growth for AR glasses. Product managers need to actively register, log in, and experience space computing, AR glasses, and agent frameworks. Just reading others' sharing reports or looking at screenshots still leaves a big gap.

If product managers cannot integrate the latest technological products and functional operations with good user experience, they cannot be good product managers.

For example, while we are still using VLA vision models for robot training, others are already using world models, which are more accurate and obviously less costly. You don't need to send your robots out to collect data, which is time - consuming and labor - intensive.

That's all for today's sharing.

This article is from the WeChat official account "Kevin's Little Bits of Changing the World" (ID: Kevingbsjddd), author: Kevin's Stories. It is published by 36Kr with authorization.