Vibe Coding = New Product Management Modern frontier models (e.g., strong coding agents like Claude) let non-coders or rusty developers describe an app idea in natural language (“vibe”), iterate via feedback, and have the AI scaffold, code, test, debug, and deploy an entire working application. English is effectively the hottest new programming language. Anyone can now go from idea → taste → product without writing code themselves. This democratizes app creation → expect a massive explosion of applications, especially filling long-tail niches that were previously uneconomical.
Training & Tuning Models = New Coding While “vibe coding” opens product creation to everyone, the highest-leverage programming today is training/tuning/fine-tuning the models themselves. This is a new form of programming: instead of hand-writing precise instructions, you design architectures, curate/tokenize massive human-generated datasets, tune hyperparameters, and search for emergent programs inside the weights. It’s programming through data and search rather than explicit logic. Domain-specific models (biology, CAD, video, sensors, games, etc.) will proliferate.
Traditional Software Engineering Is Not Dead — It’s Hyper-Leveraged AI tools make mistakes, produce leaky abstractions, suboptimal architecture, and bugs — especially outside their training distribution. Engineers who deeply understand code can debug, optimize performance, catch architectural issues, and use AI tools far more effectively than non-technical “vibe coders.” The best engineers become even more powerful. Winner-Takes-(Almost)-All + Long Tail Explosion
- No demand for average: in a world of infinite apps, only the best in each category wins big.
- More shots on goal → more niches filled, more “best” options appear.
- Head: a few giant aggregator platforms / super-apps become even larger and more dominant.
- Tail: enormous long tail of hyper-specific tiny apps.
- Medium-sized software companies get squeezed hardest.
Broader Philosophical & Practical Takes
- AI adapts to humans faster than we adapt to it.
- Entrepreneurs aren’t worried about losing jobs — the goal was never “a job” anyway.
- AIs aren’t alive and fail the ultimate intelligence test (creating new fundamental knowledge from scratch).
- Early AI adopters gain enormous, compounding edge.
- AI meets you exactly where you are — the more you know/bring, the more value you extract.
- Always leverage the best available intelligence (don’t be loyal to mediocre models/tools).
- If you can’t clearly define something, you can’t effectively program/AI it.
- Solution to AI anxiety is action — build, experiment, do. Armchair philosophizing without market/physics feedback turns hollow. Naval emphasizes he’s building again at a startup called Impossible because doing beats commenting.