Source: Anish@a16z
Fifteen years ago, if you asked, “What do smart people do on weekends?” a great answer would be, “Make YouTube videos.”
YouTube celebrated its 20th anniversary earlier this year. While we all love reminiscing about those early videos that made us laugh, we don’t always realize how countercultured it was to become a YouTuber in the early days. Even after the YouTube Partner Program launched in 2007, the idea of making money from a channel seemed unattainable for a time. With so much video content already in the world, from large studios to the long tail of content on the 110th channel in a cable TV package: how could there possibly be more commercial demand?
Looking back now, we realize that the world 15 years ago was actually the world of short videos. And we know this because of what YouTube has been doing since then. When anyone with a camera and editing tools could find their audience, we discovered a wide variety of channels and businesses—from Hot Ones to Mr Beast to Dwarkesh—that clearly deserved a prominent place in our content world. The so-called “long tail” is much larger than anyone imagined.
 Perhaps this is the right historical guide to thinking about the Large Language Model (LLM), web applications, and the future of the internet. YouTube revolutionized content by simplifying “creatively making content” and “running a small business” into a series of simple steps. You still need creativity and motivation, but the rest becomes much simpler. So why shouldn’t we see the same thing happen with software? The web has always excelled at facilitating creation by anyone without permission. But it wasn’t until the advent of the Large Language Model (LLM) that the definition of “anyone” shifted from “developers” to “anyone with an idea and access to a code agent.” Five years from now, we might look back and realize that the world lacked software because the only people capable of building software were engineers. In other words, for the rest of the internet, this is the YouTube moment: The argument that “the world lacks software” is equivalent to “the world lacks content” in 2006. In 2006, you would point to 100 cable TV channels and say, “That’s enough.” Today's software and websites face the same dilemma. Large Language Models (LLMs) are finally making it possible to develop niche software and applications that were previously unmarketable. You wouldn't hire a team of engineers to develop a product for 100 people, but you can leverage application generation and programming tools to develop (and monetize!) smaller products. Long-tail web applications will be built by specific types of "professionals," and YouTubers are the best template for such professionals we have. 
Content is now application; application is now content
Paul Bakaus points out that when people talk about the internet, they are actually talking about three different things: the content layer of the internet, including websites like blogs, YouTube, Substack, and traditional publishers; the business layer of the internet, including marketplaces like Amazon and Shopify; and the application layer of the internet, which for most of the internet's history has consisted of "serious" cloud-based software such as enterprise platforms and social networks.
Large Language Models (LLMs) influence all of these layers in different ways. The business layer is the subject of another article, but we do see LLMs playing a significant role in product recommendation and purchasing; this is evident not only in new AI discovery mechanisms like search, but more importantly in recommendation engines themselves. But putting business aside, our understanding of "content" on the internet and how it is rewarded is undergoing a major transformation.
 Content has always been the "long tail" of internet engagement, dating back 30 years, from the birth of the first web browsers to the present day. Every decade, a major story emerges revolving around the long tail of content being controlled by a centralized force: first AOL, then Facebook. This time, the biggest story is that LLM crawling sites represent a new, and even worse, form of content capture—because they've essentially become applications that let you consume information that should have been driving traffic to content creators in the past. Publishers are devastated by a phenomenon known as "Google Zero" (the day organic search traffic approaches zero), and some interesting new forces are fighting back, such as Cloudflare's foray into pay-per-view crawling and new micro-payment standards like x402. Meanwhile, something interesting is happening on the other end of the internet: Applications are becoming the new content. In the long run, the internet will become more participatory. There are some software categories that have never been developed for simple reasons: insufficient return on investment, excessive costs, or because the preferences of approximately 20 million developers dictate the software we all use. But we'll soon find out exactly what they are. 

Source: SK Ventures
Now, with new app generation tools like Replit, v0, Loveable, Figma Make, Bolt, and Base44, you can easily create prototypes, build, and release brand-new apps. Previously, this required engineering expertise, and might even have required an entire development team. Now, all you need is $200 a month (or even less!) and a good idea.
There's a famous joke that most Gen Zers aspire to grow up to be professional YouTubers or TikTokers. People love to argue and criticize whether this is a realistic goal, but they ignore a more fundamental observation: kids aspire to be entrepreneurs at the forefront of the internet, to achieve the American Dream. And to date, the main channels to achieve this aspiration are YouTube and TikTok.
 LLM offers a completely new perspective for creative professionals on the internet. If you're passionate about building your app ideas and turning them into reality, now you can. In the previous internet era, you needed to invest huge upfront costs, which were prohibitively expensive to realize an idea, and you needed to constantly grow your customer base to prove your existence. With LLM, you can easily launch your product and quickly acquire paying users. For example, my wife recently became fascinated with manifestation (a practice based on mindfulness and a positive mindset). She's incredibly good at it (so much so that I'm now extra careful not to upset her). She's now developing an online app for friends to learn manifestation. A few years ago, she might have advertised the service on Facebook and struggled to make the content engaging and dynamic. Now, she can develop an app that connects directly with customers. I'm sure this isn't just happening in my family: I've seen all sorts of apps, from traffic-controlled streaming websites in Brazil's welfare vacuum to AI band improvisation platforms built with Replit. Replit's power is evident: earlier this year, Replit's ARR reached $150 million, and demand for its coding proxy products is soaring. We can also expect to see this phenomenon extend beyond the web. Companies like Wabi are making it easier to build entirely new personal mobile applications that encompass features like weightlifting, clipart generation, fasting, and reminders to touch the grass. Making Apps: A New Kind of Entertainment A few years ago, Nadia Asparouhova partnered with Stripe Press to publish a brilliant book called *Open Work*. It tells a fascinating story of how open-source software development changed in the GitHub era: maintainers of popular projects had to spend a significant amount of time managing fans who wanted to participate and contribute. Open creation began to become a kind of "performance." Simply doing good work wasn't enough—you had to think like a broadcaster. Today, with the long tail effect of Vibe coding apps and software, we may see something similar playing out on an even larger scale. The challenges may differ, but the skill set required may be similar: successful individuals will be a certain type of professional. They will have a different intuition than previous "professional" software developers, and on how to build something that resonates. It's worth humbly remembering that few veterans of traditional television have become major stars on YouTube. Mainstream television programs, such as late-night comedies, have long viewed YouTube as a dumping ground for "extra content," such as bonus features or cuts. Traditional software developers and investors could very well end up misunderstanding vibe coding applications in the same way—not entirely dismissing them, but ignoring their purpose. As we've seen on YouTube, the most successful new apps are likely driven by individuals and personalities: these people have existing distribution channels, have created natural Schelling points for their communities, and they produce "unboxing videos, but software." But platforms like Wabi also point to another possible outcome. We might see hyper-personalized applications targeting even smaller, more niche audiences online. This would be incredibly liberating: software no longer needs to be practical. It doesn't need billions of dollars in revenue to justify its existence. All it takes is a good idea and some people with the professional intuition to understand its value and appreciate the success it requires (like professional YouTubers).