How can today's AI product managers maintain the competitiveness of their products

Currently, large AI models have almost replaced programmers of all levels, from ordinary to advanced, in software development and coding testing. The only thing AI hasn't been able to do is create a world model that can understand the world.

This is the product of spatial computing.

In the current AI Internet era, for AI product managers to keep their AI products competitive, I believe there are four core technologies.

1. Continuously dig into user needs

Good products are based on user experience, not technology. Any technology serves humanity. Therefore, the key for AI product managers is to ensure that their products best match user needs. They need to continuously dig, discover the blocking points, pain points, and pleasant points in the process of users using AI products, which requires product managers to use the products themselves constantly.

I've seen many product managers who are in charge of a certain module or product line in the company but are not the mainstream users of the product. This situation is very common in the product management industry. For example, I once worked on a stock trading software, but in fact, I don't trade stocks at all and have no interest in it. I was just doing this job for a salary or to become a product manager.

For me, I regarded that product management job as an entry - level opportunity. Some jobs require experience in large companies, while others require internship experience from school. As someone without product management work experience, this was just an opportunity to cross the entry threshold.

If a product manager is never a user of their company's product and is forced to open the product at work every day, I don't think the AI products they develop will be successful.

Secondly, not only product managers but also the core members of the entire team, including developers, product personnel, and UI designers, should be users of their own products to comprehensively dig into user needs.

It's very difficult to achieve this, but it's easier for startup teams.

So, an excellent product R & D team and startup team should first like to use their own products every day.

2. Model indexing, model mechanism, and shallow model optimization behind the shell

Except for model companies, almost all current AI products are shell - wrapped, but the ways of shell - wrapping are different. Some teams stick to using one model for shell - wrapping, while others, like manus, replace multiple models with each other. They use the most suitable model for a specific scenario instead of being locked by a single model.

This is a typical shell - wrapping thinking. When developing AI products, we should have this kind of thinking. Products in both single - modality and multi - modality should not rely on just one model company. Instead, they should provide answers or content that ordinary models can't offer based on users' input needs and the current context of the environment.

If a product only relies on one model for shell - wrapping, its ceiling will be very low, and it's easy to be eliminated by the most commonly used model company once it starts making money.

Shell - wrapping seems simple but is actually not easy.

Many AI product managers want to do it but find it difficult. At the same time, it's also something that is both difficult and easy to do. First of all, any AI product company needs a budget. Whether it's local computing power or cloud API, a large amount of testing, debugging, and evaluation by product managers are required before opening it to users. Unlike traditional product development, where we only need to check if the page path of a function is correct, pure native AI products need to generate correct answers for main scenarios, and this debugging process costs money.

Currently, there are at least four or five mainstream API providers in China. So, when to switch models and how to use the best model capabilities require spending money during the testing and usage process. This is difficult for AI product managers to achieve, but it's the easiest if the boss agrees.

Finally, there is testing and refinement.

3. Burn money before cultivating users' paid - subscription habits

In the Chinese market, the paid - subscription business model doesn't work well. It's better to turn AI into a commodity so that users are willing to pay.

For example, users pay for songs or videos. These are cases where AI is turned into virtual products, but it's difficult to achieve the cost and ROI before users subscribe.

User traffic, product user threshold, and user payment conversion require the efforts of product managers, operators, and advertising budget investment.

AI product managers also need to understand operations, the basic mechanism of traffic, and the tendency of various platform recommendation algorithms because after their AI products are launched, they need to be promoted on various platforms.

4. Really commit to lifelong learning: Keep track of model benchmarks and their advantages and disadvantages

Compared with the previous three points, this is something that AI product managers must do. Keeping track of benchmarks, changes in the general capabilities of models, the length of context parameters, and token costs can directly reduce the cost of AI products. At the same time, various agent frameworks can assist in creating their own AI products.

So, current AI products and AI product managers have a more tiring job than before. It's not only because they need to work fast, but also because the requirements come not only from users but also from the need to keep up with the latest model benchmarks. They need to understand operations, products, and agents.

This is the competitiveness of AI product managers in the current AI product field.

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), written by "Kevin's Stories", and is published by 36Kr with authorization.