Posts

Opportunity Costs It’s never easy to discover that a product you’ve poured your heart, sweat, and tears into isn’t working out. Startups operate in constant ambiguity, and sometimes you can’t see light at the end of the tunnel after toiling away for what feels like an eternity. Sometimes there simply is no light. I’ve heard the phrase “Take more market risk when you are young, and more execution risk when you are older.” As I understand it, this suggests that people early in their careers should bet on markets and opportunities, even contrarian ones. I’m reflecting on this because I’ve been thinking deeply about how to best allocate my time and energy on the most promising projects. This isn’t about diversifying—I recognize my limited attention span, and pursuing everything simultaneously leads to burnout and mediocre results. Hence this post: an attempt to provide clarity.

Every now and then I come across some article or discussion that just feels plain and mundane. All the words seem to make sense, yet at the same time, they feel almost predictable. Despite how well articulated these ideas were - be it in carefully formatted slide decks or confidently delivered proses - they fail to amaze. Ever since November of 2022, the ability to articulate words cohesively (I’m purposefully not using the word coherently) has become table stakes. In a society where frankly most work is evaluated on completion and length, LLMs have led to a rapid advancement of productivity. Yet I think we should make certain clarifications here - productivity gain is in automating repetitive and redundant tasks, this does not apply to all tasks, in fact, using GPT for sophisticated reasoning is almost guarenteed to produce mediocore results.

When you build a product with 10k or 100k users, it’s a matter of skill. But when you are planning to build a product with 1 million or 10 million users, luck is a factor that one cannot neglect. During my short time of working at ByteDance (Gauth, TikTok), I caught glimpses of what the early days of the company looked like. However, as the company grew, the early product culture slowly diluted. What does it feel to have worked at such a company during the early days, and how does one go about finding such a company? This is the question that I’ve been repeatedly thinking about recently, and trying to figure out through writing this post.

Introduction We live in an era of unprecedented access to information. The web contains almost all the knowledge needed to complete virtually any task, yet many of us still struggle to learn effectively.Our ability to ask the right questions has become the limiting factor in unlocking knowledge acquisition. This fundamental shift is transforming how we learn, build expertise, and might revolutionize education itself. The Traditional Knowledge Landscape Historically, human conversations have been the default method of acquiring knowledge. We seek out doctors for medical advice, mechanics for car problems, and teachers for academic subjects. These experts are valuable not just for their knowledge, but for their ability to understand questions we may not be able to formulate ourselves.

In a world dominated by expert opinions and algorithm-driven content, there’s something fundamentally human about wanting to know what others think. Whether we admit it or not, we’re drawn to understand the collective mindset. There’s wisdom in crowds. While large groups may not always converge on absolute truths (in fact, many truthful views begin as contrarian positions), they provide something equally valuable: comfort and context. Being part of a group, understanding its thoughts and values, creates a sense of safety and belonging that’s deeply wired into our social nature. Even when we disagree with mainstream opinions, understanding them helps us navigate social landscapes and provides reference points for our own thinking. This isn’t mere conformity—it’s about contextualizing our experiences within the broader human narrative.

The Evolution of AI Value The first wave of generative AI focused primarily on content creation - ChatGPT writing articles, Midjourney generating images, essentially replacing traditional production roles. However, as these technologies mature, their greatest value might well shift towards distribution and personalization rather than raw production. From RSS to Recommender Systems The evolution of content distribution reveals how technology repeatedly transforms information access. RSS (Really Simple Syndication) represented an early attempt to solve content discovery, providing a pull-based system where users subscribed to feeds they cared about.

Tobi Lutke’s Shopify internal memo Smaller and More Efficient Teams When should we hire a person versus delegating to AI? Recently I’ve been more reluctant towards hiring people as an attempt to build mid sized projects. Yes, the codebases would get pretty big, and there’s also tasks involved that I wouldn’t say are my forte exactly. Yet, when I think about the meetings I have to sit through communicating what I want to build and just time spent doing filler work, I get more and more inclined towards just doing it myself. It’s not to say that teamwork isn’t good work, some of the most creative product ideas I’ve worked on stemmed from chats, during lunch breaks, exploring tangents, with engineers, journalists. The value of connecting the dots during these conversations is something that is difficult to replace. However, is the assumption that delegating work means higher productivity still valid? After all, the cost of execution is continually decreasing (as long as we have a clear idea of what to build).

Here’s a list of articles that I founding interesting. I’ve attached the original article / transcript for easy refernece as well. Gary Tan on Manus: The New General-Purpose AI Agent Video URL: https://www.youtube.com/watch?v=JOYSDqJdiro Usable AI agents are finally here from deep research platforms out of OpenAI and Google to similar tools from XAI and DeepSeek. Joining the competition now is Manus, a brand new agentic AI platform that has taken the world by storm.

Hayao Miyazaki is my favorite artist and director. Though I haven’t watched his entire filmography, every moment of the films I have seen captivates me. In “Spirited Away,” the dust ball creatures exemplify his artistic prowess—his ability to infuse life into the mundane through his drawings. His work is truly a labor of love. Every scene in his animated films is hand-drawn and painted with watercolor. To put this dedication in perspective: a single 4 second crowd scene from Studio Ghibli required 1 year and 3 months to complete. At 24 frames per second, that’s 96 images—roughly 6.4 images per month or one-third of an image in an eight-hour workday. At this rate, animators would spend a decade creating just 28.8 seconds of footage. This extraordinary commitment to craft has established Miyazaki’s work as iconic for decades.

When you want to look for user pain points, go to a bar in the middle of the week, look for the most desparate looking person and buy them a beer. They’d probably have something substantial to talk about. While I don’t usually go to bars during weekdays, I’ve found some similar ways of identifying user needs, mostly through conversations or online forums. This is a running document of the user needs or pain points that I’ve found to be interesting. Perhaps some of them will turn out to be viable business oppportunities.

Year in Review 2024 felt to have passed by very quickly, in part because I had interesting work to occupy my time. Early in the year, after expanding Cogno and gaining some traction, work slowly stalled: though we did code and talk to customers, the cycles were far and wide in between. The focus on multi-agent systems was (as of writing this) in the right direction, yet we did not find the niche to tackle sales conversion improvement.

Every now and then I like to read about the advice of others who’ve succeeded in their field. Here’s a few that I personally found to be enlightening and practical. Patrick Collison: Go deep on things. Become an expert. In particular, try to go deep on multiple things. (To varying degrees, I tried to go deep on languages, programming, writing, physics, math. Some of those stuck more than others.) One of the main things you should try to achieve by age 20 is some sense for which kinds of things you enjoy doing. This probably won’t change a lot throughout your life and so you should try to discover the shape of that space as quickly as you can.

This week, I wanted to organize my thoughts about AI-generated content (AIGC) and creativity-related products from the past few months. Rather than focusing solely on my own projects, I’d like to explore the foundational aspects of AI product design, interspersing examples from my recent work. First, I want to emphasize that technology is merely a tool intended to better serve business needs. If it doesn’t significantly improve efficiency, traditional methods may be more appropriate. Second, despite the many imaginative possibilities of current technology, applications should ultimately be guided by user needs. Finally, AI technologies and markets evolve rapidly, making predictions difficult to validate, but exploring content understanding and generation remains an intriguing challenge.