# Season Execution FY2 S4

# Engineering

Decided to stop using Linear and to start using Notion for tracking app versions. 

[Version Log FY2 S4](Season%20Execution%20FY2%20S4/Version%20Log%20FY2%20S4%202a6075ee55be4c79b819cd586500c00e.csv)

# Others

- Reference Study Targets
    - 2.6: Wikipedia, Feedly, Pocket Casts
        
        Considered: Amazon Kindle, Apple Podcasts, Libby, Namu.Wiki, Spotify. 
        
    - 2.9: Instagram Threads, YouTube, Pinterest
    - 3.0: Reddit, Microsoft Teams, Microsoft Outlook
    - After that: Shopify, Web Browsers
- Introspection on Centum
    - Discovery Page
        
        Discovery Page is like visiting a mall. Enters without an intent to think about any specific topics but can discover. 
        
- Technical Challenges
    1. Managing Topics
    2. Generating Posts
    3. Curating Lists
- Distribution
    - Homebodies are likely on YouTube, Instagram, Pinterest.
    - Consider posting podcast episodes using Centum contents since homebodies listen to podcasts during chores.
    - Positioning: To listen, go to Spotify. To watch, go to YouTube. To read, come to Centum.
- Appendix: Ideas and Notes
    - Possible Technology: Offline Internet
        
        Internet is a repository containing all information available to humanity. 
        
        One trait that makes AI models technologically remarkable is that they can encode a vast amount of information into a small file size. For example, a small foundation model we are currently using for Centum is approximately 5GB which can fit trivially into consumer electronic devices, and it can fetch and form information on any topic in response to the context on command. 
        
        This is an exciting evolution for humanity. Each human at any location, even in the deep sea in the middle of the Pacific Ocean, even on another planet, even suspended in outer space, can now have and use all the information that anyone has ever known. A new age of exploration is ready to begin. 
        
        - Offline Internet Applications
            
            While offline, person can get information from offline catalog or prepare actions to be carried out when online. When online, AI companion can carry out planned actions, and save information for next time the person goes offline. 
            
            - Media
                
                While offline, access entire Wikipedia, Spotify, YouTube. 
                
            - Shopping
                
                While offline, discover and assess products, then add to cart. Place order when online again. 
                
            - Trip Planning
                
                Study cities, flights, hotels, landmarks, restaurants. Plan itinerary. Make reservations when online again. 
                
            
            We can even imagine offline community features. See Centum Community section for ideas related to community features. 
            
    - Target Customers: Travelers
        
        Travelers can benefit greatly from Centum. 
        
        Travelers need to learn about cities, flights, hotels, foods, landmarks, and souvenirs, among other topics. 
        
        Travelers can also benefit from having an “offline internet” since they may have sparse to absent connection to the online internet. 
        
    - Korean culture contents
        
        Like Eataly but for Korean culture and in an app. 
        

# FY2 S4 Season Review

Cancelled Centum for Each for now. For now, it makes more sense to fill the entire site with just contents for one kind of shoppers, and this will make it possible for us to satisfy shoppers even before an AI has had time to learn about them, making the output superior to Centum for Each from the perspective of visitors who are our target customers, at the cost of lower coverage over addressable customers. We need not grow beyond Homebody Shoppers for a long time though. 

Centum for Each was cancelled in favor of Generative Catalog. It was a dreamy experiment, but ended up feeling quite right. Specifically, adding Related Posts and Related Lists made the site feel pretty fun. 

Distribution remains an important challenge, but even if we are able to relive past seasons, we will have done work the same way in this regard, because we must not distribute widely a product that is not working well in a small scale. The product is getting there. But I note that distribution, when we get to focus on it, will present many challenges which we must overcome, thereby giving us an opportunity to grow by learning important new skills. 

Monetization also remains an important challenge. We have been diligent about planning and adding appropriate monetizations to the product this entire year, but have had trouble getting meaningful income. As for subscription, $1 a day is a high cost to consumers who are familiar with paying $0 for most apps and sites. As for ads, Google has been refusing to supply ads in Google Ads inventory to us despite our repeated efforts to form a partnership with them. The challenges inspired us to explore an interesting new opportunity in FY3. 

Finally, we tested one more dreamy experiment at the very end of FY2 S4, and ended up making a radical decision to abandon the application part of Centum into which we have been investing engineering time for multiple seasons. We will keep the database part of Centum, but switch the application part to Valencia which is a Generative Interface. Centum, like apps and sites of existing big technology corporations, starts from human-made code, then increases proportion of AI-made code. Valencia starts from AI-made code, then increases proportion of human-made code to add control. As we end FY2 S4 and enter FY3 S1, we are still learning the best way to use and profit from this new software development paradigm, but one thing that is for sure is that we will have lots of interesting learnings next season.