Fabricated Knowledge

Fabricated Knowledge

Share this post

Fabricated Knowledge
Fabricated Knowledge
AI is Creating Peak Software, Media is the Best Analogy

AI is Creating Peak Software, Media is the Best Analogy

Doug O'Laughlin
Jul 10, 2025
∙ Paid
31

Share this post

Fabricated Knowledge
Fabricated Knowledge
AI is Creating Peak Software, Media is the Best Analogy
8
1
Share

Let’s start with the basics. Software and the cost of creating software are dropping massively, as tools like Cursor and Claude Code proliferate. What used to cost thousands of dollars in tokens now costs mere cents, as generative AI is making coding much cheaper and faster.

That is an obvious statement. But so what? I think today, I want to propose my theory on how I see the broader software ecosystem will be impacted by the influx of new supply from coding agents. Welcome to the Software disruption flood. We have a very recent case study at hand to consider how software will be impacted. Today, I want to compare the rise of coding agents to the rise of YouTube and the destruction of linear media. I think it is the single best frame to view how software will change from here. So let’s set the scene.

The Internet, YouTube, and the Bundle

Once upon a time, there was a paradigm of content in which for a single subscription, you could consume 1000s of channels of content in a bundle via TV. The anchor of this bundle was Live Sports (still is), and this was the last meta of Media consumption. This lasted for decades as TV was the killer application (over radio!) from the 1980s to the mid-2010s.

Now Media itself is just content, and it can be viewed on TV in the Bundle, over-the-top (like Netflix), or even now via short-form content on TikTok, Reels, and YouTube Shorts. The key inflection point that screwed the Bundle from it’s peak of ~87% penetration to ~40s penetration today.

US traditional pay TV household penetration 2015-2023
Source: https://nscreenmedia.com/traditional-pay-tv-us-home-penetration-2023/

This is the ugly peak of TV, and compared to the very glorious rise in penetration, it’s a painful sight to watch in real time.

Meanwhile, YouTube (my proxy of choice this time) blew up in conjunction. The timing is quite accurate, and you can see the beginning of the S-curve of YouTube almost perfectly align with the top of the cable bundle. In the 2010s, as YouTube experienced explosive growth, the cable bundle reached its peak and then began to decline.

YouTube Annual Users
Source: https://prioridata.com/data/youtube-valuation/

This is what happens when your platform peaks, and I think that the Web 2.0 SaaS era bears a striking resemblance to this analogy. But before we get there, let’s talk about what killed the bundle. In hindsight, it’s pretty straightforward to attribute the issue broadly to the internet. However, I think that it wasn’t just the distribution (via the Internet) but also the media itself that changed, shifting from being concentrated in relatively few media makers (thousands of channels) to millions of individual creators. There are ~113.9 million channels on YouTube, with 32,300 with over 1 million subscribers.

For example, Mr Beast now has 400 million subscribers. That’s larger than the entire United States and much larger than Cable TV ever was. And it’s not like Mr Beast entered the ring with massive funding and a distribution channel. He raced to the top because the cost to enter the Media creation race dropped. This, to me, is the key part of the two-pronged success story of the millions of smaller creators that applies so directly to software today.

The Cost to Write Software is Fraction of the Previous Cost

I believe this is the critical insight here and why the Media disruption case makes a lot of sense compared to Software today. In 2000, it cost actual money to start a TV show. My extremely imprecise estimate is that it costs approximately $250,000 to launch a TV show in 2000, and now it costs roughly $3,000 to launch a YouTube channel.

This is likely similar to the cost-down trend that’s happening in coding assistants today. It used to cost $100s of dollars in man-hours to write ~100s of lines of code. Now it costs 100s of dollars a day to write millions of lines of code. The cost is now plummeting to enter, so I expect massive entrants everywhere.

Now, skeptics of this argument will contend that service, product market fit, and distribution (i.e., sales) are the significant differentiators. And unlike the two-pronged change in distribution and content costs for traditional media, it’s only the cost of creation that's decreasing, not the distribution costs that are changing. And frankly, I find this to be a very mediocre argument. Systems of Record are unlikely to be changed, but new business opportunities are available. And if context windows become infinite, what is the point of 1 specific solution over another if there is endless recallability and manipulation of the data and information within an LLM?

Software has always been a high-margin product, with gross margins of 90% or higher. Generative AI costs a lot, but for the creation of a product, this inherently lowers the net cost of a traditional software solution. So, if you make a lot of money, don’t have a clear differentiation from the 10s of other solutions in the market, what helps you grow? Why sales and marketing? B2B SaaS sales is a meme because it is the key differentiator between two similar products. And assuming that the bar is rising, the S&M spend “moat” feels like a race to the bottom. Now, everyone has a point solution that is ready to be sold, and this supply of software solutions will only ever go one way for the rest of our lives: Up massively.

Peak Linear TV will be an Accelerated Peak Software.

The era of making software for massive profits is over. The game used to be focused on hoarding a relatively small number of people who were able to generate the best code, thereby increasing your internal velocity and denying this supply to your competitors. Coding agents mean that the supply of software is going up exponentially, and I think that isn’t good for the entire industry.

Software feels a lot like traditional media, but will probably experience the rush of disruption much faster. Supply will flood traditional software makers, and niche solutions will eventually overwhelm the incumbents. Folks you’re looking at peak SaaS.

I think the timeline matches quite well. I believe that there is money to be made in the stocks for traditional TV. Still, it will come via a wave of consolidation as Microsoft, Salesforce, ServiceNow, and Adobe are the Disney’s and Foxes of tomorrow. The long tail is probably dead, and Cursor and other coding agents right now are adding as much revenue as the entire software space is this year. That’s your YouTube vs Bundle analogy, and I think it’s 2010. It worked out for a minute, but they have underperformed broad indexes since 2010.

chart

Software was a Local Minimum Anyway

Now, this is something I wanted to discuss anyway. If you go back in the history of computing, most software was sold as a bundle to the user anyway. Initially, software was included with the machine you purchased. IBM’s original product came free with the hardware. This changed in 1969 when IBM unbundled its software and hardware solutions.

The concept of writing software and charging money for the IP is novel, but it was only valuable when there were barriers to entry to creating that software. This was the law of the land when hardware (PCs with Intel CPUs) were ubiquitous and commoditized, and now that era of hardware is over.

In a world of infinite software generation, what is the point of software anyway? It just is an output of the hardware. In a world of hardware generating software, I don’t believe there is much of a moat to be had. The fact that original devices, such as Teradyne’s ATEs, early CNCs, Switches, and Cray computers, were custom software solutions for their intended tasks, bundled into the devices, makes me think that we could return to that direction.

Software was just a local minimum on the actual march of progress, and that was hardware all along. I want to reiterate this: one of the most magical things we have ever done is to impose information using physics onto silicon chips. Everything else is downstream, and now we have created a generalizable model that can create, organize, and generate information from a sufficiently large number of hardware transistors. Software is likely to resemble a local minimum in the long term as the value shifts back to the hardware. And as Claude code and cursor increase the supply of software, we go back to the real scarcity, hardware.

Some further thoughts behind the paywall.

This post is for paid subscribers

Already a paid subscriber? Sign in
© 2025 Doug OLaughlin
Privacy ∙ Terms ∙ Collection notice
Start writingGet the app
Substack is the home for great culture

Share