On Artificial Intelligence — shantum's notes

ELI5 / TLDR

Naval argues that AI is best understood not as a thinking creature but as the latest layer in the programming abstraction stack — one that lets anyone “code” in plain English. The real winners are people with software engineering intuition (they can fix what the AI gets wrong) and entrepreneurs (they have agency the AI lacks). He thinks AI anxiety is mostly ignorance, creativity remains a distinctly human trick, and the correct response to all of it is to open the hood and learn how the machine actually works.

The Full Story

Vibe Coding and the Long Tail of Apps

Naval’s starting observation is blunt: tools like Claude Code have turned English into a programming language. You describe what you want, give feedback, and a working application comes out the other end — no code written by hand.

Vibe coding is the new product management. Training and tuning models is the new coding.

The consequence is a tsunami of applications. But Naval doesn’t think that means a tsunami of success. Markets are winner-take-all — nobody wants the second-best app. What changes is that far more niches get filled. An app for tracking lunar phases in a specific context, a nostalgia-tuned video game, a personal health tracker shaped exactly to your needs — these couldn’t justify an engineer’s salary before, but they can justify a weekend of vibe coding.

The pattern mirrors what Amazon did to bookstores and YouTube did to TV: one massive aggregator at the top, a handful of dominant apps, and then a vast long tail of tiny creations. The mid-sized software companies — the 5-to-20 person shops filling enterprise niches — are the ones that get crushed.

The New Programmers

If vibe coding handles the surface, then what counts as real programming now? Naval says it’s training and tuning models. Traditional coding gives the computer precise step-by-step instructions. AI training is the inverse: you pour massive datasets into a structure you’ve designed, and the system searches for a program that can reproduce or extend that data.

You’ve set up a model. You’ve tuned the number of parameters. You’ve tuned the learning rate. You’ve tuned the batch size. You’re pouring it inside the system you’ve designed almost like a giant pachinko machine.

The output is a program that handles fuzzy, real-world problems — creative writing, image generation, web search — where “correct” has many acceptable answers. These AI researchers are the new elite programmers, and they’re paid accordingly.

But traditional software engineers aren’t dead. They still have two advantages: they think in code (so they can fix what AI gets wrong, since all abstractions leak), and they operate at the edges of what’s possible, outside the AI’s training distribution. High-performance code, novel architectures, genuinely new problems — these still need a human hand.

Become the best in the world at what you do. Keep redefining what you do until this is true.

Why Entrepreneurs Sleep Fine

Naval makes a clean distinction: a job is a slot someone else defined. Entrepreneurship is self-directed problem-solving in unknown territory. AI can’t replace the latter because AI has no agency — no desires, no survival instinct, no skin in the game. An AI that helps you build your product is an ally, not a competitor.

The key thing that distinguishes entrepreneurs from everybody else right now in the economy is entrepreneurs have extreme agency.

The same logic applies to scientists, explorers, and true artists. If your work is so difficult and self-directed that you welcome any tool that helps, AI is pure upside. It doesn’t steal your job because you don’t have a job — you have a problem you’re trying to solve.

The Photography Analogy

Art used to push toward realism. Then photography arrived and handled realism perfectly, so art got weird — surrealism, abstraction, modernism. Some of the greatest creativity emerged precisely because the mechanical part was automated.

AI is doing the same thing to software, writing, and creative work generally. The basic thing becomes trivially easy to produce. A few will stand out by creating something genuinely novel on top of that baseline. And it would be hard to argue that society is worse off for having photography, even though portrait painters lost their livelihood.

AI as the Patient Tutor

Naval calls AI the most powerful learning tool that has ever existed — not because it knows more than a textbook, but because it meets you at your exact level. Eighth-grade vocabulary but fifth-grade math? It adjusts. You don’t feel stupid, and you don’t feel bored.

The means of learning are abundant. It’s the desire to learn that’s scarce.

He runs every query through four models, cross-checks them against each other, and drills into visual diagrams and analogies. In technical or scientific domains — where political bias doesn’t distort the output — he finds them close to flawless.

Creativity, Consciousness, and the Limits

Naval is skeptical that current AI is creative in any meaningful sense. Solving an unsolved math problem isn’t creativity if the answer was effectively embedded across the training data in five places, three forms, and two languages. The AI assembled it — impressive, but mechanistic.

Creativity is much more in the domain of coming up with an answer that was not predictable or foreseeable from the question and from the elements that were already known.

He also doesn’t buy superintelligence panic. Humans are “universal explainers” — anything possible under physics, we can eventually model in our heads. A calculator is already superhuman at arithmetic, but nobody calls it superintelligent. AI will be superhuman at many tasks and incompetent at others, just like every machine ever built.

Steve Jobs famously said that a computer is a bicycle for the mind. Maybe now we have a motorcycle for the mind. But you still need someone to ride it.

The Antidote to AI Anxiety

Naval’s closing advice: if AI makes you anxious, the cure is action, not avoidance. Open the hood. Learn how it works — not to the level where you can build it, but to your own satisfaction. Understanding the machine does two things: it makes you better at using it, and it lets you calibrate your fear. Is this Skynet, or is this a very good tool with clear boundaries?

The solution to anxiety is always action. Anxiety is a non-specific fear that things are going to go poorly and your brain and body are telling you to do something about it, but you’re not sure what.

Claude’s Take

This is Naval at his sharpest — grounded in the daily reality of building software rather than floating in abstraction. The podcast format (walking around, no script) gives it a loose quality that actually helps. He’s not performing; he’s thinking out loud.

The strongest insight is the one about competitive intelligence being zero-sum. If every guy has an AI earpiece on a date, every woman gets one too, and the remaining edge is entirely human. That’s a clean, memorable way to cut through the hype about AI replacing everything.

Where he’s weaker is on creativity. His definition — something unpredictable from the inputs — is philosophically tidy but practically hard to apply. Plenty of human “creativity” is just recombination from a larger, messier training set (lived experience). The line between AI recombination and human recombination may be blurrier than he suggests.

The photography analogy is genuinely useful and I haven’t heard it deployed this well in an AI context before. It reframes the disruption question: not “will AI destroy X?” but “what weird, interesting new directions will emerge once the mechanical part is automated?”

Score: 8/10. Dense, practical, wide-ranging. Naval’s builder credibility (he’s actively coding with these tools, not just commenting) elevates it above the average AI discourse. Loses points only for some repetition in the middle section and a few ideas that have been in his rotation for years.