The "one-person company" is bullshit. It looks at what AI can do today and assumes nothing else will change. Sure, AI lets 2 people do the work of 10 — but that's against today's demands. As tools get better, they create new needs. What 2 people can handle today won't be enough 5 years from now. There's also a simpler problem: people like working together. If everyone runs a one-person company, sooner or later two of them will team up — and instantly beat every solo competitor. That's not a maybe. That's just game theory.
Claude Code can do anything that OpenClaw can do with the right tools. OpenClaw is nothing fancy. But its UX integrates I/O into commonly used channels and lowers the barrier so that everyday users can see the power of AI agents, instead of it being just a developer tool like Claude Code. Powered by AI, more and more apps will be revolutionized in terms of interaction, not just become smarter or faster.
What I learned from @bcherny: build products for 6 months later. You can bet that: 1. tokens will be more affordable 2. models will be more intelligent
Build tools for AI models and agents.
It makes more sense that AI interacts with another AI and minimizes the interaction with humans.
Tried Atlas. Honestly, not much better than Comet or other AI browsers. The real problem is the prompt-based interface—it feels clunky and unnatural. AI in browsers makes sense, but users shouldn’t have to “tell” it what to do. I am imagining an interface that just gets what you need and guides you effortlessly.
When you're holding a stock or token, you should be hungry for information, resources, and knowledge about it. If you're not actively seeking to understand your investment, you probably shouldn't be holding it. Deep research is critical—and AI has made this easier than ever. I've always had great respect for people who knew how to find what they needed on the internet, even before AI existed. At the foundation of all this is reading: it's the essential skill that makes everything else possible.
Some non-trivial stuff in modern software development: types, states, workflows/agents with generative models. AI is not to be generating tedious reports for humans but to speed up the interactions between agents/ programs. Now that we have generative UI, would there be something like generative types?
Recognizing the "anti-fragile" aspect of something is the first step toward making progress. This means acknowledging that you're not perfect and will make mistakes, but you try anyway, accept feedback, reflect on it, learn from it, and begin the next cycle. However, if you don't act because you fear making mistakes—which is "fragile"—you'll never move forward.
Thoughts after watching Hassan's talk: to build a successful product, MAKE IT SUPER GOOD-LOOKING, AND EASY TO UNDERSTAND, EASY TO USE, AND EASY TO SHARE.