samesh

Composition is all you need. Watch the full video below.

I'm going to say something shocking and borderline heretical (for me). I'm thinking about switching from Python to TypeScript for my next project. I've been using Python for *decades* as my primary programming language. There are so many things I love about the language: 1) Python is a joy to develop in. 2) Great libraries for everything I need to do. 3) It seems to be lingua franca of AI. It's what LLMs use when they need to generate code to solve a problem. It's usually the first language to get an SDK/library from the frontier model companies (and others in the space). It's what many AI-oriented open source projects are built-in. So, why am I considering TypeScript? A few reasons: 1) There's elegance and value in having both my front-end code and back-end code in the same language. 2) TypeScript is natively type-safe. 3) When distributing applications to others (particularly CLI tools), it's much more common/simple to do with a Node app then try to build binaries in Python (using something like pyinstall). 4) TypeScript is a close second when it comes to being popular in the AI community. 5) There are packages for most of the common things I need (web framework, database access, web/http calls, etc.). 6) There is first-class support from Vercel for deploying TypeScript apps. I'm both an investor in Vercel and a customer but have mostly used it for front-end deploys, not backend. And, what once was the biggest reason NOT to use TypeScript is no longer true: The fact that I don't know TypeScript and didn't want to spend hundreds of hours becoming an expert at it. Now with agentic coding, I don't need deep knowledge of the language in order to be productive. With my knowledge of Python, C++ and other prior languages, I can likely get by pretty well in TypeScript with the help of Codex and Claude Code. Haven't made the decision to switch over completely yet, but next time I have a small, contained project I need to work on, I'm considering trying TypeScript. What do you think? Am I overthinking it or underthinking it?

Vibe coding is more addictive than any video game ever made (if you know what you want to build).

Startup survival rates haven't changed in 30 years. ~80% survive year one, ~50% at five, ~35% at ten. Same numbers in 1994 as today. Through dot-com, cloud, mobile, ZIRP, AI. Every input changed. The output didn't. Most people read this as a failure of methodology. Lean Startup is in 97% of university entrepreneurship curricula. Didn't move the needle. The usual conclusion is we need better methods, or the methods get competed away once everyone adopts them. I think the chart isn't measuring what we think it's measuring. Progress has a dependency structure. Every innovation sits in a graph with prerequisites — technologies, infrastructure, behavioral norms that must exist before the new thing becomes feasible. Webvan raised $375M for grocery delivery in 1999 and went bankrupt. Not because the idea was bad. Because smartphones, GPS tracking, gig labor, and consumer trust in strangers at your door didn't exist yet. Instacart worked fifteen years later because the prerequisites had resolved. Webvan was probing a position in the graph that looked real but wasn't. A phantom node. The graph branches. Each unlock exposes new feasible nodes. Personal computing exposes operating systems, which expose the web, which exposes social, which exposes mobile-native, which exposes gig economy, which exposes delivery. The frontier gets wider every generation. Over 30 years, the number of startups grew and average size at birth shrank. More probes, cheaper probes. If the number of real nodes were fixed, survival rates should have dropped. But they didn't. Because the frontier expands proportionally. More unlocks means more visible opportunities means more founders showing up. The entrants don't arrive independently of the frontier growing. They arrive because it grew. The chart is flat because it's measuring the shape of the opportunity frontier, not founder quality. Methods like Lean Startup improve how you search within a feasible node. Competition determines who wins a ripe node. The dependency graph determines which nodes are feasible at all. Methods and competition decide who succeeds. The graph decides how many opportunities are real. Which makes "visionary founder" a concrete skill: graph-reading. The ability to look at the current state of the world and see which nodes are about to become real. Serial founders repeat because this transfers across domains. The internet, mobile, cloud, and AI each unlocked entire new layers of the graph. The branching factor exploded every time. Survival rate still didn't move. More nodes, proportionally more founders. The graph gets bigger but its shape stays the same. The flat chart isn't a failure. It's a fingerprint.

seeing 18 yos try to grind startups is heartbreaking to me. when i was that age i would spend all day in the college library finding obscure books nobody will ever read. not a single care about what i will do with that knowledge. easily the most valuable time of my life