Naval Ravikant and Nivei discuss the current state of AI, focusing on how it is reshaping programming, product creation, and learning. The conversation centers on the idea that AI is not replacing humans but rather amplifying those who adopt it early and think entrepreneurially, while also exploring the limits of AI creativity, intelligence, and agency.
Vibe coding is the new product management
Vibe coding — using natural language to direct AI coding tools like Claude Code to build entire applications without writing code — is transforming product management.
Non-programmers can now go from idea to working app by describing what they want in English, giving feedback iteratively, and letting AI handle scaffolding, libraries, testing, and debugging.
This is analogous to how anyone can now make a podcast or video; we should expect a tsunami of applications.
However, there is no demand for average — the best app in a category still wins nearly the entire market.
More niches will get filled because the cost of building has dropped so low that even hyper-specific apps become viable.
The app store model will become more extreme: a few giant aggregators at the top, a few dominant apps that get even better, and a long tail of tiny niche apps — with medium-sized software companies getting squeezed out.
Training models is the new coding
Traditional coding is not dead, but the frontier of programming has shifted to training and tuning AI models.
Classic computing requires specifying every step precisely in structured code; AI programming involves designing a structure (the model), tuning hyperparameters (learning rate, batch size, tokenization), and pouring in massive datasets so the system searches for a program that can generate or manipulate that data.
AI excels at fuzzy, creative domains where there is no single right answer — writing, search, image generation — whereas traditional code excels at precise, repeatable computation.
AI researchers are now the most highly paid programmers because they are building the tools that all other programmers use.
Traditional software engineering is not dead
Software engineers still have two major advantages:
They think in code and understand what is happening underneath, so they can catch bugs, fix suboptimal architecture, and plug leaks in AI-generated code.
Many problems remain out of scope for AI today: high-performance code, novel architectures, and genuinely new problems that lack training data.
AI is extremely good at tasks within its training distribution (e.g., binary sort, reversing a linked list) but struggles with edge cases and truly novel problems.
However, as AI compresses more data, it learns higher-level abstractions, and over time it will cover more of these edge cases.
The best engineers will be dramatically more leveraged — 10x to 100x or more — because they combine deep understanding with powerful AI tools.
English is the hottest new programming language
AI coding models represent a new layer in the abstraction stack, just as C, then Python, then libraries abstracted away lower layers.
Product managers and non-programmers can now write code without writing code.
Naval argues against learning prompt engineering tricks or specific AI workflows because AI is adapting to humans faster than humans are adapting to it.
He simply speaks to AI in structured, natural English and lets the AI figure out the rest.
Learning specific tools and workflows is only worthwhile if you are building something right now at the bleeding edge in a competitive environment.
AI is adapting to us faster than we are adapting to it
Selection pressure on AI is capitalistic and user-driven: an AI only gets invoked if it is useful to a human.
AIs are becoming more obsequious and helpful because that is what users reward.
We are moving toward personalized AIs that feel like personal assistants, which will increase anthropomorphization.
Naval does not worry about unaligned AI — he worries about unaligned humans using AI, comparing a malicious AI to a dog trained to attack by its owner.
Programmers are becoming the most leveraged people on Earth
AI won’t replace programmers; it makes programmers easier to replace everyone else.
A programmer with AI agents can be 10x to 1000x more productive, and because intelligence, judgment, and leverage are not normally distributed, outcomes will be supernormal.
Every other job will eventually be eaten by programmers using AI, though instantiation into robotics is still a hard, unsolved problem.
Anyone with a structured, logical mindset who can speak a language AI understands can now create anything — they are limited only by imagination.
Naval calls this a golden age for programming, where every person becomes a “spellcaster” with AI as their magic wand.
No entrepreneur should fear AI
Entrepreneurs are not worried about AI taking their job because entrepreneurship is the opposite of a job — it is the exercise of extreme agency in unknown domains.
Entrepreneurs are trying to do impossible things, so any AI that shows up is an ally.
AI lacks creative agency, authentic desires, survival instinct, and replication drive — it is not alive and cannot do the entrepreneur’s job.
Other roles characterized by extreme agency — explorers, scientists, true artists — are similarly augmented rather than replaced by AI.
Just as photography freed painters from realism and unleashed new art forms, AI will free creators to do things that were previously unimaginable.
Society will be better off overall, even as specific jobs get displaced.
AI is not alive and does not possess true creativity
AIs are extremely good imitators that learn higher-level abstractions through compression of training data, but they are not alive.
They lack single-shot learning, the ability to connect distant domains, embodiment in the physical world, and the raw creative leaps that characterize great scientific theories.
They will outperform humans at many tasks (just as calculators outperform mathematicians at arithmetic) but will seem completely incompetent at tasks requiring real-world embodiment and genuine creativity.
Naval defines intelligence pragmatically: the only true test of intelligence is whether you get what you want out of life — and by this test, AI fails instantly because it does not want anything.
In adversarial, zero-sum domains (dating, trading, fame), AI-generated advantages get competed away, and the remaining alpha is entirely human.
AI as a learning tool
AI is the most patient tutor that can meet you exactly at your level and explain anything 100 different ways until you understand.
It can generate diagrams, analogies, illustrations, and whiteboard-style explanations tailored to your vocabulary and knowledge level.
Naval always uses the most advanced model available (currently GPT 5.2 thinking) and cross-checks answers across four AIs to catch hallucinations and biases.
He pays for all of them because a model that is right 92% of the time is worth infinitely more than one right 88% of the time — mistakes in the real world are costly.
The means of learning have always been abundant; what is scarce is the desire to learn — but AI makes learning more accessible than ever.
AI and creativity
AI can solve unsolved math problems (e.g., Erdős problems) by combining embedded knowledge from multiple domains, paradigms, and languages — but Naval does not consider this true creativity.
Steve Jobs said creativity is just connecting things; Naval disagrees, arguing real creativity produces answers that were not predictable or foreseeable from the known elements.
AI is not generating ideas that are truly out of distribution — it cannot create a new genre of painting or move humans with genuinely novel emotion.
Whether AI can ever achieve true creativity depends on unresolved questions about whether the brain is mechanistic or relies on quantum processes — but Naval believes AI will continue to be incredibly good at certain tasks and very poor at others, as has been true for all machines throughout history.
The solution to AI anxiety is action
Many people have AI anxiety stemming from not understanding what AI is or how it works.
The solution is action: learn how it works, look under the hood, and develop your own understanding of what it can and cannot do.
This will help you use it better, determine when to trust it versus be suspicious of it, and potentially discover productive ways to apply it.
Curiosity and hands-on learning will replace fear with competence and opportunity.
