For five hundred years, publishing has depended on a simple economic principle: write once, sell many times.

A publisher often incurs high fixed costs to create a work, but recovers that expense with wide distribution at low marginal costs. In short, you pay the author and the editor, design the product, then print and reprint as many times as the market will support.

Digital distribution put that model on steroids by cutting distribution costs substantially. Even “new media” like podcasts and videos use the same model. That is, there’s still only one product that’s distributed to every customer at a low marginal cost.

AI is changing that.

Samantha and The AI Inversion

AI agents introduce a fundamentally different dynamic because you no longer assume that the same product will be delivered to every customer. The AI can generate custom content for each user. (Like “Samantha” from “Her.”)

I recently experimented with this idea by building an engine for a custom podcast. I like to hear what’s going on in the world while I cook my daily omelet, and I like to hear from multiple sources, with some pushback on the controversial bits.

ChatGPT and I designed a script to collect the news feeds I was interested in, pick the particular stories that match my interests, summarize them, provide some pushback, assemble it all into a podcast script, and then send the script to a text-to-speech service to create my daily download.

The result was great, but it also used a tremendous number of tokens. Doing it every day would be expensive, and there doesn’t seem to be an easy way to convert the concept into a commercial venture. At least not yet.

Still, the experiment revealed something subtle but important.

AI personalization reintroduces marginal cost into publishing

In traditional publishing, the marginal cost of distributing digital content is close to zero. AI-generated digital content is even cheaper because you eliminate some of the fixed costs as well.

Personalized AI content (like my custom podcast idea) reintroduces marginal cost. Each computational step requires tokens that ding the balance sheet.

This transforms publishing from a manufacturing model into a service model. Instead of producing a fixed object and distributing it widely, the publisher (or the platform) now performs a series of computations for each individual user to create a customized product.

That product could be a podcast, a training video, a personal development plan, or a financial to-do list – all customized to the specific requirements of the end user.

Personalization Isn’t Just a Feature — It’s a Cost Structure

These personalized products change the nature of publishing in several ways. First, let’s address cost.

1. Content

Until publishers wake up to this new reality and block the bots, the content is mostly free. For my proof-of-concept podcast idea, I found several free RSS feeds to get my daily news. In the future, publishers will have to gate that content.

2. Processing

Gathering and processing the content uses up AI resources. Some of that can be trimmed by siphoning off some of the work to internal resources, but right now it’s still a costly endeavor.

3. Delivery

Digital delivery remains inexpensive, but personalized content doesn’t have to be digital.

What happens when users expect …

• Customized children’s books
• Individually assembled legal briefings
• Personalized print magazines
• Educational materials tailored to their exact requirements

Print-on-demand and custom assembly reintroduce physical variability — and with it, real-world fulfillment costs.

Mass publishing sought to eliminate variability. AI will bring it back.

Where do publishers fit in?

Let’s go back to my daily podcast. Right now, if I want to listen to a podcast, the publisher monetizes it with ads. When I’m using AI to summarize their news feed, the publisher gets nothing.

With my podcast concept, I’m not buying my daily podcast from a publisher. My AI agent is assembling it from several different publishers’ feeds.
My agent can summarize ten sources, compare viewpoints, and present a custom synthesis in seconds. A fixed, static article written for somebody else doesn’t compare.

The Economic Paradox

AI lowers the cost of producing generic content. But it may raise the cost of producing personalized content at scale. That’s the paradox.

For centuries, scale was the advantage. The whole point was to amortize fixed costs across as many copies as possible.

But in an AI-agent world, the “copy” itself becomes unstable. There is no single edition. There are infinite micro-editions.

In some sense, we may be returning — digitally — to a pre-print model. Before the printing press, every manuscript was copied by hand. Every version was slightly different. Every book was, in effect, a custom artifact.

The printing press standardized content and drove down marginal costs. AI may fragment it again.

Three Emerging Business Models

If this shift is real, publishers will need to choose their posture.

1. The Centralized AI Publisher

In this model, the publisher owns the AI stack. It generates personalized content internally and charges subscription fees that cover computational costs.
This is capital-intensive. Margins will depend heavily on optimizing which tools to use when. The publisher becomes part media company, part infrastructure provider.

2. The Publisher as Expert

In this scenario, the publisher provides structured content, trusted data, and authoritative source material to the AI. The user’s AI agent handles personalization. Here, the publisher becomes a knowledge supplier — a trusted source rather than a direct content assembler.

The competitive advantage shifts from format to authority.

3. The Hybrid Model

The core content remains standardized. AI layers personalization on top — summaries, alternate angles, format shifts, contextual adaptations.
Write once. Interpret many times.

This preserves some of the old economics while acknowledging the new expectation of customization.

A Larger Question

My customized podcast experiment was exciting and I believe we’re heading in that general direction. It also raises some deeper questions.

• Is extreme personalization economically sustainable?
• Is it culturally desirable?

Mass publishing created shared experiences. Millions read the same article. Millions watched the same broadcast. Culture had common reference points.

If every reader receives a different version, optimized for their interests and biases, what happens to shared discourse and human interaction? Samantha (in “Her”) started monopolizing Theodore’s time.

Perhaps the more important question for publishers is this:

In a world where AI agents can generate perfectly tailored content for every individual, what is the enduring value of their product.

For five hundred years, the answer was scale.

In the age of AI agents, the answer may be something else entirely.