Naval Ravikant on Vibe Coding, Training Models, and the New App Economics
Naval Ravikant and Nivi on the state of AI-assisted coding in February 2026 — focusing on vibe coding as a new form of product management, training models as the new coding discipline, and the structural economics of the coming app store explosion.
Key ideas
- Vibe coding is the new product management. AI coding agents translate English directly into working software; product managers who once managed engineers now manage AI agents. The leverage profile of PM work has been transferred to anyone with taste and a prompt.
- Training models is the new coding. Classic coding specifies every step formally; training a model is a different kind of programming — designing a structure and pouring data through it to find a program. The people doing this are the new coders.
- App economics: no demand for average. A tsunami of AI-generated apps is coming. The best app for any use case will still win the category; the middle tier of 5–20-person software companies serving enterprise niches will be disrupted; but a long tail of niche apps will flourish, and the platform aggregators will capture super-wealth.
- AI as bicycle for the mind. Steve Jobs’ metaphor updated: AI is a motorcycle for the mind — enormously amplifying throughput for those who know how to ride it, negligible for those who don’t.
- Impossible and the doing imperative. Naval has returned to building (at Impossible) because he believes commentators without feedback from physics and free markets become armchair philosophers. Grounding in work keeps thinking honest.
Vibe Coding as New Product Management
The shift Naval describes: Claude Code and similar agents allow anyone to describe an application, get it laid out, built, tested, and iterated on — without writing a single line of code. The role of product manager (describe what the product should do; manage a team to build it) has been absorbed into the vibe coding loop. The PM’s skill — having taste, knowing what to build, giving feedback — now directs AI agents directly.
Naval’s implication: the people who were traditionally bottlenecked by their inability to code can now express their taste directly as product.
Training as Programming
Naval distinguishes classic computing (specify every step formally; computer is a deterministic calculator) from AI training (design a structure; pour data through it; the system finds a program). Both are programming disciplines, but training requires different skills: intuition for data quality, model architecture, hyperparameter tuning. The people doing this — the ML researchers and engineers — are the new programmers for the most important software in the world.
App Store Economics
Naval predicts a structural outcome:
- Best-of-category apps continue to win entire categories — the head gets bigger.
- Middle-tier SaaS (5–20 person companies serving a specific enterprise niche) gets disrupted — their niche can now be vibe-coded away.
- Long-tail niche apps explode — markets that were too small for a full engineering team now get served.
- Aggregators (the app stores themselves) capture super-wealth, as they do whenever content multiplies.