Ein weiterer berühmter Gründer betritt das Feld der AI-Podcasts, und Sequoia China setzt darauf. Wird diesmal etwas passieren?
In our previous topics, we have observed several cases of "star entrepreneurs" entering the field of AI podcasts, including ChatPods by Zhang Yueguang and Laifu by Jiao Ke.
The downloads of ChatPods and Laifu on both platforms in the last 90 days | Image source: Diandian Data
After a certain period of time, the data for both products doesn't look very good. According to Diandian Data, the global download volume of ChatPods in September was 35,000. Due to the small scale of the product, third - party data couldn't capture active users, and the monthly revenue was less than $100. For Laifu, which was launched later, the data was even worse. In September, only about 2,000 downloads were registered, and no MAU and revenue data could be captured.
The work experience of Ethan KJ Li | Image source: LinkedIn
The setbacks of the first two products haven't dampened the enthusiasm of the follow - up entrepreneurs. Recently, Ethan KJ Li, the former CEO of the Intelligent Education Unit of ByteDance, has also entered the field of "AI podcasts", but with a different approach from the first two.
Ethan KJ Li's product is called Aibrary. It was launched for testing in the US App Store on April 23 this year and officially launched on September 23. According to the description on the official website, the core function of Aibrary is to transform/re - design books into personalized podcasts and provide customized learning paths and interactive consultations for personal learning situations. Before the product was launched, its parent company Ouraca had already completed several rounds of financing. The investors include well - known venture capital firms such as Sequoia Capital China, CXC Capital, and Factorial Fund. With the background of a "star entrepreneur + well - known VC", Aibrary is, so to speak, "born with a silver spoon in its mouth".
As the previous analysis has shown, it's not very promising for either ChatPods, which wants to use AI capabilities to make podcast listening more efficient, or products like Laifu, which wants to directly generate AI podcasts to compete with human hosts. However, the approach of using AI podcasts for the knowledge acquisition scenario has proven effective in products like NotebookLM. How will Ethan KJ Li's attempt differ from star products like NotebookLM?
1. This is the best - designed AI podcast product we've ever tested
Overall, Aibrary is mainly targeted at personal learning and self - improvement situations. Compared with other products like NotebookLM and Speechify that we've observed, there are two core differences in Aibrary: firstly, a relatively comprehensive recommendation system and content system; secondly, the use of AI podcasts as the main carrier for content.
The recommended content on Aibrary's homepage (Picture 1, 2, 3), the second main entry Nova (Picture 4)
When entering Aibrary, users need to complete a 6 - step registration process. In addition to the usual information such as age, gender, and interesting topics, users are also asked to select the people they admire and the goals for using Aibrary. This selection then determines the recommended content on the homepage (Home). According to co - founder Wu Jundong, Aibrary has developed a recommendation algorithm to recommend content according to user preferences. In addition, the homepage also provides a search function. Users can search for books by title or interesting topics as keywords and then learn further.
In addition to the above - mentioned homepage with recommendations and search, Aibrary also has the main entry Chatbot "Nova" and the user profile. In "Nova", users can communicate with the chatbot (as shown in Picture 4) and put forward more personalized requirements. Aibrary's chatbot has three different roles: Nova, which mainly interacts with users; Orion, which provides knowledge management services (questions about books, building knowledge graphs, etc.); and Atlas, which is mainly responsible for goal decomposition and user review.
From my experience, most answers are given by Nova. Only for some more complex questions do Orion and Atlas offer their views, but their participation is not very high. We still don't fully understand why the task of the common context is taken over by 3 chatbots.
For more complex answers, Orion and Atlas are involved
Whether it's homepage recommendations, search results, or chatbot Q&A, all are divided into 3 categories: book lists that may meet user needs; details of specific books selected by the user (including a text description, an abstract & 2 audio files of the podcast, as well as a purchase link for the book); and the system - generated idea Twin Podcast when the user wants to learn more about a specific topic in a selected book.
Book lists (Picture 1, 2), book details (Picture 3) | Image source: Aibrary
The details of a selected book are also not what we traditionally understand as an e - book, but "two audio files + purchase link" (as shown in the right picture above). Both audio files last about 8 - 10 minutes. The first is a "monologue" summary, and the second is a "dialogue" podcast. The first is an introduction to the book's outline, and the second analyzes the outline in the form of a podcast. Both audio files are provided by the system.
Regarding copyright, since Aibrary only generates summaries and podcasts based on the book content and doesn't use the original book cover, legal experts say that this is in a legal gray area, and the copyright risk is low both in China and abroad.
After users listen to the two audio files, they get a first impression of the book. If they are interested, they can either click on the lower purchase link to jump to the Amazon website and buy the paper book, or click on "Make idea Twin Podcast" to generate an even more personalized podcast with AI and understand the book content more deeply.
Book selection (Picture 1), host voice selection (Picture 2), user information and voice cloning (Picture 3, 4)
The AI - generated podcast is called "Idea Twin Podcast", and it's also the biggest difference between Aibrary and other similar products that we've noticed.
The Idea Twin Podcast is a real - time - generated two - host podcast. The two participants are the AI host and the "user clone". Users first need to select a book and the voice of the AI host, then enter personal information such as educational background, major, occupation, MBTI, personal description, etc., and clone their own voice. The system then generates a "clone" of the user based on the entered information, which communicates with the AI host as another role. Generating an Idea Twin Podcast costs 100 credits. Non - subscribed users can only generate one podcast.
The Idea Twin Podcast follows a pattern where the host (AI voice) asks questions and the "clone" (user voice) gives answers. Users can think about the questions asked by the AI while listening to the Q&A and follow the answers of their "own" voice to promote active thinking ability.
According to the developers: "With Aibrary, you don't just listen to more podcasts, but co - host a show with AI." From my experience, the answers of the "clone" actually take into account the entered personal information, and the user's own voice enhances the sense of immersion, but the sense of immersion is still a bit superficial at present. While listening, I still feel that it's a conversation between two AIs, and it's difficult to put myself in the role of the "clone".
The relationship between the functions and content of Aibrary | Created by Baijing Chuhai
In our previous topics, we've observed several "AI + audio podcast" products. We've created a quadrant diagram to clarify the positioning of different "AI + podcast" products based on the purpose (knowledge acquisition/creation tool) and the way of AI integration (generation/understanding). | Created by Baijing Chuhai
From the two diagrams of product design and industry positioning, it can be seen that Aibrary has actually found a relatively unoccupied position. Earlier products like Doubao have difficulties processing long books into short podcasts. Compared with Speechify, which reads long books word - for - word, Aibrary offers a "from short to long" guiding experience and lowers the threshold for "reading" with podcasts.
What's described above is essentially the phase in which the product introduces users into the learning process. The functions such as setting learning paths and interactive consultations, which are mentioned on the official website and in the podcast and are supposed to continuously monitor users in achieving their learning goals, are not very noticeable. Currently, the project still gives the impression of an "unfinished" state. But as a product that was officially launched at the end of last year, it already has a certain degree of differentiation and a reasonable path. Compared with earlier products, it has already shown a certain attractiveness. Behind this is the in - depth thinking of some founders who have been in the education industry for many years.
2. Why have the star founders who have been engaged in K12 education for a long time chosen the lifelong learning sector?
The founder of Aibrary, Ethan KJ Li, previously developed the K12 education collaboration platform "Jike Big Data" for schools and educational institutions, which was later acquired by ByteDance. He then joined ByteDance as the CEO of the Intelligent Education Unit and left the company in 2022. After that, he was briefly a partner at CXC Capital and founded the company Ouraca in Silicon Valley in January 2025. The other two co - founders have also been in the education industry for a long time. Among them, Wu Jundong is the former director of international investment and strategic development of TAL Education Group, and Zhang Qiming is the former head of the education middleware of ByteDance.
From the resumes of the three founders, it can be seen that they were all once "pioneers" in the Chinese education industry. However, in the AI era, they didn't continue in the K12 sector but instead focused on the relatively unknown sector of "lifelong learning for adults". This change stems from their thinking about learning in the AI era.
The founder, Ethan KJ Li, writes in an article:
"In the not - too - distant future, AI can answer almost all questions. Therefore, the "answers" will gradually become less important, while the "questions" and the underlying thinking process have higher value. Therefore, education in the AI era must change from "content dissemination" to "cognitive restructuring". If we follow this thinking, it's particularly important to get users to think independently.
Ander...