New podcast on vibe coding - A Return to Code. A Return to Coding 00:20 The Personal App Store 03:17 Vibe Coding Is a Video Game with Real-World Rewards 06:22 Pure Software Is Uninvestable 10:33 A Place for Each Model 14:22 AI Is Eager to Please 17:57 Why Math... and Coding? 22:10 The Beginning of the End of Apple’s Dominance 24:17 Coding Agents As Customer Service Reps 27:55show more

Apple: Spotify: YouTube: Transcript: — Presented by AngelList:

@naval I’ll gift you a free lifetime @joinladder membership so you can spend more time vibe‑coding the rest of your personal app store... But we’re just honored Ladder made the cut as inspo 🤝

@joinladder I’m actually a ladder subscriber :-) But sometimes I want something a lot simpler, so I vibe coded the app on the side.

@joinladder @naval dreams do come true... We should have done a campaign with you and Hilary Duff to break the internet

For Naval with Love & Squalor:

Is this AI generated? Voice sounds robotic. But agree, vibe coding is fun. I hope compute gets faster though as lots if time is wasted waiting

No AI.

Palantir trades north of 80x sales with zero hardware. The highest-multiple corner of the public market right now is pure software with a real data moat. That chapter title might be a 2014 take aging into 2026.

Actually listen to the podcast instead of reacting.

Listened, fair. Palantir fits your investable bucket (data moat plus embedded gov distribution). The pure software no-moat point holds

I've never downloaded a podcast so fast.

This is very much my experience with vibecoding over the last couple of months. I have extensive and deep knowledge in networking and security architecture, but I'm not a programmer by any means. I can articulate fairly well what I want the system to do and it's unbelievable the systems I've been able to build.

been thinking about the personal app store part-feels like we're heading back to everyone having their own weird utilities instead of one-size-fits-all saas

"Pure software is uninvestable" is the chapter I'm most curious about. Naval calling the end of Apple's dominance in the same podcast is a lot to unpack. Listening now

@gkisokay The world hates vibe coders. Naval cooking or cooked ?

if yo uare founder or want to be then this podcast is really true gem 💎

Naval, your case against Apple assumes frontier intelligence wins in consumer. I disagree. The bottleneck in consumer AI is context, distribution, and price. Apple owns all three. Frontier intelligence doesn't matter for 99% of consumer use cases. The "agent generates your interface on the fly" future doesn't require GPT-6 or Claude Opus 5. It requires a model good enough to summarize your texts, draft a reply, find a photo, set a reminder, and hand off to a specialist when needed. We cleared that bar 6 months ago. OpenAI and Anthropic can burn trillions racing for state-of-the-art, but consumers don't have the purchasing power or the desire to live on the frontier. They want something good enough, private, and cheap or free. Frontier capability matters for enterprise and developers. The person checking their calendar doesn't care. Open-source plus post-training has already collapsed the model moat. The Cursor playbook is now well understood. Take open weights, post-train for your specific use case, ship something competitive with frontier labs at a fraction of the cost. Apple doesn't need to spend $100B on training runs. They can sit six months behind the frontier, post-train for on-device inference, and capture 99% of the consumer value at near-zero marginal cost. The hundreds of billions they've "saved" by underdelivering on their own AI bet looks less like a failure and more like patience once you accept that frontier capability was never the moat. The real bottleneck is ambient context ingestion, and Apple owns that layer. When models commoditize, the next bottleneck becomes context. What does the agent actually know about you, your day, your environment, your intent? That's a sensor problem. Core Motion, HealthKit, HomeKit, the accessibility APIs, screen content via ScreenCaptureKit, App Intents, Live Activities, Focus modes, location and significant location changes, on-device messages and mail, calendar and contacts, the microphone and camera, the LiDAR scanner, the U1 ultra-wideband chip for spatial awareness, and the Secure Enclave for processing all of it privately. Apple has spent 15 years building the most sophisticated context-ingest layer in consumer technology. They have a device in 1.5 billion pockets that already knows when you're driving, who you're meeting, what you just looked at, what room of the house you're in, and what your heart rate is doing. Whoever owns the context layer wins the next decade for consumer AI, and only Apple has the hardware footprint, privacy story, and OS-level integration to do it at scale. Pricing gravity favors Apple. Most consumers won't pay more than $10/month for software. A huge share won't pay anything. The labs need $20-$200/month subscriptions to fund their compute bills, and reports suggest even those prices represent a 5x subsidy of true unit costs. Apple doesn't need to monetize AI. Better models just become the reason to upgrade to the latest iPhone. That asymmetry is fatal for the labs (hence why OpenAI is trying to develop their own phone). A local, private, free model that's 80% as good as the frontier kills every consumer AI subscription on the market. The labs are racing to be the best. Apple just has to be free, private, and connected to your personal context layer. I'm betting on the local, private, on-device future (at least for consumer AI). And Apple is the company most likely to deliver it.

@naval “AI is eager to please” → exactly what happened when I asked them to draw me in your style 😂 ChatGPT: good Claude: hilarious Grok: nightmare Full experiment here 👇

I will tune in Is it also available on YouTube?

If this doesn't get you building I don't know what will.

The shift from "which app do I open" to "just talk to AI" is what kills Apple's moat. Hadn't thought about it that way before, but now I can't unthink it

I don't understand why vibe-coding is thrown around as an insult.🙄 Mainly the people who spent weeks learning syntax for every new language their employer threw at them. The same people who laughed at no-code, at drag-and-drop. The future of work isn't knowing how to write a for loop. It's knowing what to build, and being courageous enough to build it.

personal appstore is the way to go.

Vibe coding became a useful frame because it forces you to talk about the gap between intent and code. Curious whether the episode pushes back on it or treats it as the new default for shipping software.

So many fascinating parallels between vibecoding and video games:

You mentioned on this podcast that you're vibe coding AirChat all over again. Any chance it's going to come back?

Interesting comparison to self-driving cars. Having worked at Waymo and now at Replit, I am often comparing these two magical technologies. But I'd never considered what you mentioned, that both of them free you from your normal inhibitions. Whether it's interacting with a driver or with an engineer, the presence of others inhibits us. We put on a mask and censor ourselves so that we don't look dumb. But when you are vibecoding on your iPhone in a self-driving car, you can be as dumb as you like. I never thought about how freeing vibecoding is in that sense. The absence of judgment is an incredibly powerful and liberating thing.

Naval always breaks down big tech shifts in the smartest way. This episode on vibe coding and AI agents sounds like must listen stuff for anyone building or creating today.

Can't wait to listen.

the 'AI is eager to please' one is worth the whole episode. That's why our agents kept producing plausible-sounding garbage. The model just agrees with whatever you seemed to want. We added a separate evaluator agent to catch it before anything ships. Works.

Vibe coding brings people in, but staying requires real understanding.

I vibe coded a vr teleop for robot data collection

the 'personal app store' bit is the framing nobody else has named. most of what i make with these is single user, lives one afternoon, never sees an install or a review queue.

the "vibe coding is a video game" angle is so spot on, you get rewarded in real products not just points

@Cod3xOrg

Vibe coding" captures something real - when the feedback loop gets fast enough, it stops feeling like work and starts feeling like play.

We are in the AI ERA. Claude can take on Google now. Regardless of that, Do SEO.

Thanks for posting!

The "AI is eager to please" part explained so much weird hallucination behavior I’ve seen. How do you personally prompt to get past that eagerness and force truth instead?

Prime take 💯🎯 There's a steep step from a good prototype to deploying a product, as it might involve rebuiding it, and most importantly - involving people in that. I guess we'll also see a lot of vibe coded products going straight to production, as people can skip this step.