In my last column, I discussed how AI paywalls dramatically outperform paywalls operated by teams of expert editors. That invited some skepticism (eg. “How does the AI know when a big story breaks?”), so let’s drill into how these new paywalls work, and let’s see if we can identify some insights beyond the media industry into technology and AI in general.
How does a system learn, evolve, and react to changing conditions in real time? Well, how do you do it? You can’t just passively receive signals. You also have to transmit some of your own.
These new paywalls are constantly monitoring and reacting to new traffic sources and patterns. But crucially, they also send out their own signals in order to learn and adapt, much like bats use echolocation.
Here’s where things get a little random.
In his book “Notes on Complexity,” Neil Theise, a professor of pathology at the NYU Grossman School of Medicine, discusses the concept of quenched disorder, or the necessary “Goldilocks” amount of randomness that enables systems to adapt and thrive.
As he puts it, “New opportunities can’t be found if random things aren’t happening here and there… A little randomness keeps the system alive.”
Without some unpredictability baked in, complex systems become brittle, over-optimized, and ultimately incapable of responding to change. Overly controlled systems tend to stagnate, while those that allow for bounded (the keyword here is bounded) experimentation create space for innovation.
This is precisely why quenched disorder is an integral function of Paywall 3.0 systems.
The Curiosity Engine
When Jonathan Harris, Senior Director of Product at Zuora, first began to test his new AI agent against traditional publisher paywalls, he made a jaw-dropping discovery.
“A completely random set of offers and blocks actually did better than the publisher’s fixed rules,” he recalls. “So we realized—we just had to be better at random.”
Let that sink in. A system that made decisions by essentially flipping a coin repeatedly outperformed paywalls refined over the years by teams of optimization specialists. Not because randomness is smart, but because any set of fixed rules is essentially backward-looking.
As a result, a portion of Paywall 3.0’s outreach effort is intentionally exploratory, or effectively governed by chance. By introducing controlled randomness (e.g., varying offers, pricing, access rules, or engagement prompts), Paywall 3.0 frameworks identify new conversion paths and remain responsive to changing audiences.
Over time, successful actions become greedy. They’re shown more often, because they work (until something better comes along). But the system never fully locks in. It’s constantly testing. It continues to try out alternatives because user behavior, market conditions, and content value never stop changing.
Too much exploration hurts short-term revenue. Too much exploitation prevents improvement. Balancing the two enables Paywall 3.0 systems to improve conversion rates and lifetime value simultaneously. A team of humans, no matter how smart and news-savvy, just can’t compete against the scientific method at scale.
Wake Up Your Database
Now imagine applying this kind of agentic optimization beyond media sites to knowledge platforms, financial databases, or any system built on open data and variable demand. You could be sitting on a gold mine.
Because at the end of the day, the only things that make your company truly valuable and truly unique are a) your customer relationships and b) your data. This is particularly relevant in the era of AI, where the whole concept of corporate sovereignty is being interrogated (more on that later!).
If you have a public database, you’re potentially sitting on a huge wealth of opportunity. You just need a curiosity engine.
