Weekly writing about how technology and people intersect. By day, I’m building Daybreak to partner with early-stage founders. By night, I’m writing Digital Native about market trends and startup opportunities.

If you haven’t subscribed, join 65,000+ weekly readers by subscribing here:

The Age of Fragmentation: AI’s Impact on Content and Code

Last week, I saw this chart on Twitter:

About 1 in 5 Americans used to share the same favorite athlete. Today, we’re down to 1 in 20. Derek Thompson shared the tweet with this takeaway: “Fun snapshot of the death of mainstream & erosion of mass-media celebrity.”

I do think that’s one interpretation: our monoculture has fragmented with the internet. Another interpretation is just that Michael Jordan was really, really good at basketball—and consequently really, really popular. Maybe he was the outlier and now we’re back to the steady-state ~5%, where we always should have been.

But let’s run with the fragmentation angle.

Cultural fragmentation has been a common topic on Digital Native—a few years back I published a piece called, “The Internet Killed Mainstream Culture.” A better title would have been some riff on “Video Killed the Radio Star” (“The Internet Killed the Mainstream Media Star?”) but you get the point. The internet fragmented culture by giving us new rails for information and content. Down with the gatekeepers, etc etc.

Just look at the declining viewership of award shows, or at the explosion of content (“peak TV”) that has made for fewer common cultural touchstones. In the mid-20th century we had three TV channels; now we have 500+ scripted shows a year.

Those charts show the downward trend of mainstream culture, but how do we quantify online fragmentation?

Take the size of YouTube as an example. Ethan Zuckerman tried to estimate just how big YouTube is through random number sampling. A typical YouTube URL looks something like this:

https://www.youtube.com/ watch?v=vXPJVwwEmiM

After the “watch?v=” comes an 11-digit string, and the last digit can only be one of 16 values. So there are 2^64 possible YouTube URLs—about 18.4 quintillion (!). Zuckerman did some “drunk dialing” to estimate how many YouTube videos there are. He explains:

Consider drunk dialing again. Let’s assume you only dial numbers in the 413 area code: 413-000-0000 through 413-999-9999. That’s 10,000,000 possible numbers. If one in 100 phone calls connect, you can estimate that 100,000 people have numbers in the 413 area code. In our case, our drunk dials tried roughly 32K numbers at the same time, and we got a “hit” every 50,000 times or so. Our current estimate for the size of YouTube is 13.325 billion videos—we are now updating this number every few weeks at tubestats.org.

That’s a lot of videos—and we’re seeing steady year-on-year compounding:

More anecdotally: I consider myself a chronically-online person, yet when I see lists of “50 Biggest Stars on YouTube,” I recognize barely half the names. Culture, meet fragmentation.

The thesis statement for this week’s piece: the internet broke open culture and software by breaking open distribution; AI will break open culture and software by breaking open production. This week’s piece is about how we think we’ve seen a lot of fragmentation for content and for code—but it’s nothing compared to what’s coming.

Examples of AI Fragmentation

As AI proliferates content and code, the long tail will get longer.

In the 2010s, a popular thesis was investing in internet niches. I can’t find the exact phrasing or source, but a common maxim went something like this: “Find a subreddit with 1M+ members, and it can be turned into a $1B+ business.” The “unbundlings” of Craigslist and Ebay were popular charts a decade ago.

The takeaway was clear: what seem like small niches can turn out to be pretty big. That’s what happens when you have 5B+ people online.

With AI, we should again be paying attention to niches: the long tail is about to get a lot longer for both culture and software. The companies that capture value will be those that underpin the niche: platforms for AI content creation and AI app development.

It all comes down to cost of production: AI lowers the cost of production—by a lot. Say I need to create a logo for Daybreak. Before AI, I’d need to hire a graphic designer and it would take them a couple hours. I’d probably shell out a few hundred bucks, maybe a few thousand. Today, there are dozens of AI tools I can turn to; for most, I’ll pay a monthly subscription and get the result within seconds.

With a lower cost of production, more content and more code gets made—stuff that wasn’t possible in the past, because the cost was prohibitive or the tools inaccessible. A film or app that was artificially gated by barriers becomes ungated.

So what’s investable here for early-stage AI? Companies that provide the (1) tooling and (2) networks for content and software.

Let’s take two examples of startups—one in culture and one in software:

Culture Example: Sekai

Sekai is one of the most fascinating AI companies I’ve come across.

Sekai—whose name means “world” or “earth” in Japanese—is a gen AI storytelling platform. Users can generate stories about, well, anything—and Sekai builds on the Story Protocol so that creators can protect and earn from the intellectual property they create. It’s a unique fusion of AI and blockchain.

You can think of Sekai as fan fiction on steroids, with new tools for content production and new rails for content ownership. Here’s what the landing page looks like:

I can prompt Sekai to tell me a story about Harry Potter’s and Draco Malfoy’s kids having adventures in Hogwarts. Within moments, I get an AI-generated story about Albus Potter and Scorpius Malfoy discovering The Tempest Stone at Hogwarts. The story has AI images and audio, complete with dialogue and narration. (Read my story here.)

This is very cool—Sekai is creating a new category. Think about Netflix pioneering a new form of storytelling with streaming video, or Roblox pioneering a new form of storytelling with Roblox Studio and game discovery. Sekai uses AI to supercharge fanfic, building off Lego-like characters and locations and IPs from various universes.

Software Example: Lovable

Sekai is a good example of AI making content creation easier: anyone can now tell a robust multimedia story with AI. This will no doubt fragment culture further. But what about software? We’re also seeing new platforms for software creation. Earlier we looked at the growth of YouTube to illustrate the growth of content; we could run a similar analysis on GitHub to show the growth of code.

What’s an example of AI software creation? Take Lovable. Lovable bills itself as your superhuman full-stack engineer.

Let’s follow the sample prompt they give us. I’ll type in this:

Can you create a landing page for my venture capital firm, Daybreak, that shows a pretty sunrise and explains that we are a thesis-driven venture capital firm focused on partnering with early-stage technology companies?

Here’s the result. On the left you can see my Lovable engineer’s “scratch work” and explanation; on the right you see the result. This took about 30 seconds to generate.

Say I want to give some feedback—I instruct Lovable to put “Daybreak” in capital letters and to add a 3D element. After a few seconds of generation, I get this:

I produced both the Sekai and Lovable outputs for free, each within a minute or two. Using AI, we can create new stories and new applications that would have previously taken hours or cost large sums of money. Stuff that wouldn’t get made now gets made. When creation is this accessible, production flourishes. And when production flourishes, we get fragmentation.

AI Production

Let’s continue with new AI tools for production. I want to illustrate just how easy it’s becoming to make stuff.

Let’s pick an image and video creation tool. How about Ideogram? I like its style.

I’ve always been obsessed with Pixar, and I recently had an idea for a Pixar story. It’s half-baked, so bear with me. The story follows the life of an anthropomorphic dollar bill, as he adventures through the world—from a homeless man’s coffers to a billionaire’s pocket. “The Secret Life of Money,” if you will. What better protagonist to tell a story about income inequality? (If you’re thinking, “This guy should stick to VC”—don’t worry, I plan to.)

Here’s a prompt I wrote for Ideogram—I even messed up the prompt by copying and pasting from another prompt, but it still worked:

Here’s the output image:

Great, we have our dollar bill protagonist. Let’s call him Billy.

We now want to animate Billy. So let’s use another free tool—Minimax, a Chinese AI video generator. I select “image-to-video” and instruct the product to give Billy movement, having him walk down the streets of New York.

Voila. This is what we get:

With a bit more work, I could begin crafting a short film—or even a feature. Making this video of Billy took me just a couple minutes. It’s never been easier to bring an idea to life.

When everyone can make their own Pixar film, what happens? When everyone can build their own app, what happens? Content and code explode. The cream still rises to the top, but we all have to sort through the explosion to find quality.

Final Thoughts: Where’s the value? Applications. Where’s the defensibility? Networks.

Note that the companies here are largely application layer companies. In the past week, it’s all of a sudden looking like this is where the defensibility lies in AI. (Not a surprise to many in venture, who have been saying this for years.) This is because the internet is buzzing with America’s Next Top Model. Not Tyra Banks—DeepSeek.

We won’t re-litigate DeepSeek here, since it’s been covered ad nauseam. Ben Thompson had a good FAQ here. I will say that the ~$5M training cost is clearly false, designed by a Chinese company to undermine U.S. tech stocks. (It worked, wiping $600B off Nvidia—the biggest loss in a day. I guarantee you the Chinese hedge fund behind DeepSeek had taken out shorts against Nvidia before releasing DeepSeek.)

The DeepSeek saga this week shows us that the model layer is a tricky place to be. Applications, meanwhile, are looking more attractive, with the model “tax” dropping precipitously.

The DeepSeek news is also a reminder of something else: building networks helps you build a moat.

Sekai not only allows me to tell stories, but allows me to invite collaborators, to publish my stories to the community, and to build with other people’s IP. All of these features add defensibility. When the inevitable competitors arise, Sekai will be fortified.

Likewise, we’ve seen numerous image creation companies emerge, each carving out its own niche: Pika seems particularly good for anime; Midjourney has its futuristic style; and so on. These companies can differentiate by building networks—places to share creations, collaborate, and so on— and we see top players starting to do this.

And if ChatGPT did a better job with social or network features, the switching cost to DeepSeek would be higher.

The next few years will see a Renaissance in creation—for both content and code. The content creation will splinter culture further; the code creation will splinter software further. The valuable companies that get built will do two things: (1) they’ll build the tooling for content creation and code creation, and (2) they’ll build the networks around those creations.

Thanks for reading! Subscribe here to receive Digital Native in your inbox each week: