Original video in Chinese.
Key Takeaway
- Third product: LearnFromYouTube, which extracts the script from YouTube videos to generate Chinese/English study reports (8 parts: summary, key points, outline, standout quotes, etc.), dissecting long videos like a master chef carving up an ox; inspired by NotebookLM and utility websites.
- Market insight: In the AI era, there are big opportunities for low-priced, single-function atomic tools; people prefer ready-made products over building their own, and traffic benefits from search engines.
- Product journey: The author’s three products (Prompt House prompt management, Obsidian MCP note operations, and this one) have been recognized (included by the official MCP project, GitHub stars); having moved from content creator to developer, AI coding is powerful and builders are reaping the dividends.
I’ve built my third product as well. You can use it now by logging into the LearnFromYouTube website.
This product is very simple. You enter a YouTube video link, and it first extracts the script, then generates a Chinese or English study report based on your choice.
This report is divided into eight sections, including a content summary, key points, an outline, standout quotes from the video, and more. If I had to describe its value in one phrase, it would be: dissecting an ox the way a master butcher does.
A YouTube video, especially a long one over half an hour, contains too much information. So I want AI to help me break it down and extract it in a framework-like structure. Just like a professional butcher, after dismembering the content, divide it into eight parts and lay them out neatly on the cutting board.
My inspiration for this product came from two places:
First, NotebookLM.
I recommended this product in several videos before. It completely reshaped the way we learn. Among its features are generating briefing documents, study guides, FAQs, and timeline reports.
I think that’s pretty useful, but too simple. So I made one myself based on my own ideas.
Second, all kinds of utility websites.
There are many utility websites that people have probably used, but never really noticed. For example, JPG to PNG, JPG to PDF, and extracting YouTube video scripts.
These utility websites are extremely unremarkable, and their functions are very single-purpose, but over time they’ve gotten a huge amount of traffic from search engines.
I quite like this kind of low-cost, quietly making a fortune approach. And I’ve said in my community before that in the AI era, the rules for search engine traffic allocation will change, and the value of utility websites will change too. For example, when I saw those websites for downloading YouTube scripts, I thought:
Besides scripts, could we use AI to provide even more valuable things?
That’s how this simple product came about.
To be honest, if you don’t mind the hassle, you could use the AI tools you already have to create a study report like this; you don’t need my product. But I’m very certain that:
As long as the price is low enough and the results are good enough, 99% of people will choose a ready-made product instead of building it themselves.
Token prices are still going down, and model performance is still going up. So I believe that beyond mainstream AI products, there will still be a large number of atomic tools built with AI technology, aimed at niche needs, single-function tools.
This is the direction I want to try next. LearnFromYouTube is just the beginning.
So, up to now, I’ve built three products.
The first is Prompt House, a prompt management and invocation tool. Its core is connecting to AI clients through MCP, allowing the AI client to autonomously choose which prompt to use.
The second is Obsidian MCP, open source, under the MIT license. In addition to operating on notes and folders, it also supports having AI help generate backlinks, and more.
When I shared these two products before, quite a few people mocked me, thinking that as someone without a technical background, what kind of product could I possibly build?
But the fact is, Prompt House has been quite successful. Beyond making dollars, what makes me even happier is that it was included by the official MCP project. You can see it in MCP’s official GitHub. That shows that Prompt House’s quality and value have already been recognized by the community.
My second product, Obsidian MCP, now has 140 stars on GitHub too. Once I have some time, I’ll add a few more features, post about it on Reddit again, and the number of stars will keep going up. And if I submit it, it will definitely also be included by the official MCP project.
Now when I go out to events and check in, I see that my identity is no longer content creator, but developer.
To be honest, as creative work goes, building products is much more enjoyable than making content. The biggest reward and biggest feedback from building products is seeing your own ideas get validated.
For example, both of my first two products were based on the same judgment: Context Matters.
I believe you either do foundational model R&D, or you provide context for the model, then package it into a product.
Enough suitable context, combined with AI’s capabilities, can produce new value.
As for the complete product development process, I shared it all within the community. AI coding is already strong enough now to completely replace the act of “writing code.” All I need to do well are the remaining parts, such as framework design. The overseas ecosystem is excellent, with lots of services like Supabase. So building products feels incredibly comfortable.
If anyone tells you today that AI coding doesn’t work, you can just block them outright. That kind of person is both stupid and arrogant, and they’re hopeless.
This wave of AI will make the division between two types of people even clearer: producers and consumers. Producers will receive the biggest technological dividends.
But the threshold for being a producer is higher than before. For example, there are a large number of programmers who can only write code and have almost no other abilities. People like that, along with a large number of white-collar workers who don’t have much ability and are just drifting through life, will be replaced by AI and kicked out of the producer ranks.
I can only wish everyone good luck. I will keep moving forward, and I will never be backward compatible.
OK, that’s it for this episode. If you want to understand AI, want to become a super individual, and want to find like-minded people, come join our newtype community. See you next time!