This article delves into the key elements of product-market matching (PMF) in AI applications, and reveals the challenges and opportunities of technical product matching (TPF) in the AI era by analyzing the cases of Cursor and Arc browsers.
I am Uncle Huang of AI products, currently working as an AI product consultant for Honor Mobile Phone and 360 companies, a super consultant of WaytoAGI, and the author of the AI Programming Blue Book. In the process of playing AI programming, there are also a lot of ideas to land, in the past few years, I have also studied and thought about products, today’s article is a recent text version of the Tsinghua AI application development training camp.
I will try to explain the key points of PMF in easy-to-understand voice, and use 4 AI applications: Cursor, Duxiang, Xinguang, and Arc Browser (this article mainly talks about 2) to do case breakdown. As Uncle Huang shifts more energy to AI applications, he will also practice making more AI applications and have in-depth exchanges with more AI application developers, including the overseas part, Learn in Public, Build in Public, and the content will continue to iterate and be organized at the appropriate timeAI Application PMF Blue Book。
What does a product manager need to do?
In the process of a product from scratch, it is not easy to do a good job in the role of product manager, in addition to the well-known writing requirements, writing requirements, writing requirements, there are many things to do. The product manager is not what you think, but will only ask you for trouble, make a request:
View details >
What is PMF?
Marc Andreessen is very aptly what he said:
1. Good market
2. Products that meet the market
Another perspective is Paul Graham’s YC creed:Make Something People Want
Further disassembly, here are three elements:
- Market demand: that is, there are really people who are willing to pay (or their time and energy)
- Market potential: There is enough market to earn revenue to support the business model
- Competitive opportunities: Opportunities for new solutions, new experiences – old experiences > migration costs
In this, we can talk about the new opportunities brought by the AI era, that is, the competitive opportunity:
In the previous article about Fellou (after Dia, Fellou is another AI browser that I highly recommend), Uncle Huang wrote that their Deep Search (note that it is not a research report type of DeepResearch) does a very good job in graphic expression, such as in the picture above, after reading this report, I can directly make a decision whether to buy a MacBook Air or a Pro, which is good in breadth and depth, and the visual presentation is excellent. It can be read in a few glances, which is far better than the smelly and long pile of text given by other AI searches.
Including a large number of pictures in this article, they are all from the leading AIPPT tool in the broken gear: Gamma (Ten Questions Gamma: How to practice the cliff AIPPT king?) Uncle Huang also wrote an analysis.
These are all good products that can use new technologies to meet old needs in the AI era.
Of course, the new variables also challenge the original product paradigm to some extent:
The challenge of AI product PMF
One of the most important points is: TPF, or TCPF
Technology-product matching, or technology+cost-product matching
The impression is that this concept proposed by Kai-Fu Lee first summarizes the new changes brought about by the new paradigm, and Uncle Huang made a summary last year:
To put it simply, the core element is to find a good problem, which is urgently needed by users and the market, and it comes from your understanding of the industry, insight into users, and forward-looking judgment of technology.
After a good question, technical verification should be carried out, including whether it can be done, how to do it, speed and cost considerations.
The product solution will also include many functional points, which can give more consideration to functional potential, confidence in obtaining good results, ease of implementation and other dimensions. These evaluation dimensions can help us do a good job in demand ranking, but to score more objectively, we need continuous practical feedback after the product is launched.
It is also because of this that it is very important to get your hands dirty, don’t sit in the office and blow the air conditioner, and don’t rely purely on your brain to spray your needs.
Let’s use a few products to expand on the above theory a bit.
Case 1: Cursor
Cursor should be known to everyone, some people say that it is just a magic modification of VS Code, but it does not prevent us from recharging it regularly every month.
It was born from an underlying belief: Vim
In an interview with Lex Friedman, they responded that we are all Vim users.
This is very important, to understand first of all, you need to know what Vim is, Vim is an almost mythical editor in the technical circle, it is loved by masters, it has risen to a belief in extreme efficiency, minimalism and low-level control.
This belief directly leads to the birth path of Cursor, which is quite typical:
Cursor does not do plugins, but directly does the underlying refactoring.
Looking back now, we can easily say: Isn’t it a magic change? But people without faith will not really do this.
Vim’s biggest inspiration for Cursor is an innovative philosophy of “returning to the essence, reconstructing the bottom, and pursuing the ultimate”. This philosophy allows Cursor not to be satisfied with being a “smarter plugin”, but to become the true masterpiece of developers in the AI era. As the history of technological progress has repeatedly demonstrated, only products that dare to challenge the bottom and reshape the paradigm have the potential to truly change the world.
Uncle Huang likes this passage very much!
The second is the evolution of AI, from Scalling Laws, which is rarely mentioned in the media, to the emergence of Github Copilot, and GPT4, the founding team of Cursor has been paying attention to and thinking about what dimensions will be changed by the continuous evolution of AI.
Remember the competitive opportunities in the PMF points we mentioned earlier? Keep in mind:
Old solutions are not good enough, and new opportunities are opening up.
All these variables finally exploded after the release of Claude 3.5 Sonnet, as the model capabilities broke through the “physical limits” in the field of AI programming:
The story of Cloudflare’s 8-year-old daughter using Cursor to build a chatbot in 45 minutes detonated Cursor.
Just as the controversy over the painting at the Space Opera detonated Midjourney, the moment has come for Cursor.
Cursor has reached PMF!
Case 2: Arc’s failure
Frankly speaking, now I am on the desktop, the main browser I use is Arc, its aesthetics, sidebar for tab management, Space design, I can’t get rid of my love, but this does not prevent Arc from failing in the mass market, Arc is a carnival belonging to a niche group!
As you can see, Josh Miller (founder of Arc) mentioned important questions in his review:
1. There is no clear problem to solve
2. Led by a small number of internal test users
3. Too many features have been added
This led to a tear between the vision and the actual needs of the users!
This is actually a serious problem that is often encountered in PMF:
PMF for small sample size users often encounters serious challenges when going to the masses!
Arc is tied up by a small number of geek users, resulting in demand that cannot meet the needs of the mass group, and the impression is that they have been in internal testing for a very long time and have been meeting the needs of a small group of people.
Arc is a very typical closed country, holding back the big move behind closed doors.
We must not make such a mistake.
The product is just PMF in the geek, but if you want to go to the public, you will naturally encounter a growth dilemma:
So although Arc won the runner-up in the 2023PH product list, it still had to make a new browser in the end:
Of course, the lessons from old failures are so profound that Dia’s design is extremely laconic, AI First.
