I have been thinking about network effects for more than a decade. Recently, I found myself thinking about them again, but this time the question feels quite different: What happens when a network's participants are a mix of humans and AI agents?

That question led me to ASSET 3.0.

A: Atomic Unit of Value
S: Synthesize the Unit of Value
S: Spark Demand
E: Exponential Value
T: Transform Intelligence

I will unpack each of these in my next post. For now, I want to focus on why I think ASSET needs another evolution, because something fundamental is changing underneath the products we are building.

The network has a new participant—the AI agent.

And once I started thinking about agents as participants, rather than simply technology that helps humans do things faster, I started seeing network effects differently.

The Network Effect We Know

When I first developed ASSET, I was thinking about what I had learned building Wattpad and observing other companies with network effects. At Wattpad, the Atomic Unit was a story. Writers contributed stories; stories attracted readers; readers commented, voted, and shared; and all that activity encouraged writers to contribute more. Some readers eventually became writers themselves. Supply attracted demand, demand created more supply, and the network became more valuable as participation grew.

At the same time, something else was accumulating underneath that human network: data.

Every time someone read a story, finished a chapter, commented on a paragraph, added something to a reading list or shared it with someone else, they created a signal. As those signals accumulated, they started to tell us which stories were gaining momentum, where readers were deeply engaged and which audiences were responding to different kinds of stories. Machine learning could use those signals to improve recommendations and discovery, and eventually that proprietary understanding of audience engagement also helped Wattpad identify stories with potential to become books, movies and television shows through Wattpad Studios.

So the human network was also creating a data network effect.

The more people participated, the more proprietary data the network generated. The richer the data became, the more useful machine learning could become, which could make the product more valuable and encourage even more participation. Human participation created proprietary data, proprietary data improved the product, and a better product attracted more human participation.

That flywheel became an important foundation for many machine learning companies.

GenAI Accelerated the Flywheel

When generative AI arrived, I updated the framework to ASSET 2.0 because GenAI changed the economics of both creation and consumption. The platform can use generative AI to seed the initial set of atomic units. A writer could create more quickly, a designer could generate images, a developer could generate code, and someone consuming information could personalize or remix what they received. The gap between consuming something and creating something shrank, so the supply-and-demand flywheel could move much faster.

That also accelerated the data network effect. More creation led to more consumption, which created more interactions and richer signals, and those signals could help AI create increasingly useful experiences. Humans and AI were collaborating throughout the network, with human activity continuing to generate the proprietary data that made the system more valuable.

Now, in this agentic AI era, agents introduce another evolution because they can increasingly understand context, reason about it, make decisions, use tools, take actions and interact with other agents and humans. That gives AI a very different role inside the network.

AI can become a participant.

From Data Network Effects to Intelligence Network Effects

Think about a customer inside a large enterprise. Sales knows the relationship with that customer, Customer Success understands how they are doing, Support knows the issues they have experienced, Product understands how they are using the product, Finance understands the economics, and Legal understands the contract. All of this knowledge is about the same customer, even though it may live across many different systems, teams and people.

Now imagine AI agents operating around that same customer. A Customer Success agent may notice that product usage has dropped significantly and, as it looks more closely at how the customer is using the product, discover that the decline started shortly after an integration stopped working. That immediately makes the Support history more relevant. A related ticket may give an Engineering agent the context it needs to understand the technical issue, while the fact that the contract is approaching renewal gives all of this information a very different level of commercial importance.

What becomes interesting is what each participant contributes back. One agent brings an observation about usage, another adds the history of the customer relationship, another contributes technical understanding, and a human may add judgment that changes what happens next. Each contribution enriches the shared understanding of the customer, making the next participant more useful and the next action more valuable.

This is where I think the data network effect starts to evolve into something richer.

In the previous era, human participation generated behavioural data. A click, purchase, comment, view, search or story created a signal, and machine learning learned from the accumulation of those signals. In an agentic network, the participants can also contribute context, reasoning, decisions, actions and outcomes. An agent can understand a situation, decide what matters, take an action and observe what happens next. A human can add judgment. Another agent can use the resulting experience as context for what it does next.

Over time, the enterprise can begin to understand something much richer than what happened. It can increasingly understand what we knew, what we decided, what we did, and what happened as a result.

That starts to look like organizational intelligence.

Imagine the customer whose usage started declining. The integration issue gets identified and fixed, usage recovers, and the customer renews. The value of that experience can travel beyond this particular customer. The context that mattered, the actions that worked and the outcome that followed can become useful to another agent working with another customer in the future.

This is the transition I find fascinating.

Human participation created data networks. Human and agent participation can create intelligence networks.

A data network becomes more valuable as participation generates more proprietary data. An intelligence network becomes more valuable as humans and agents contribute context, decisions, actions, and outcomes that enrich future work. The network begins to accumulate experience about how the organization actually gets things done.

And as that experience accumulates, the enterprise itself can become more intelligent.

The Intelligence Network Effect

This is why I think the agentic era deserves a new way of thinking about network effects. The opportunity becomes especially interesting when multiple agents and humans are working around the same Unit of Value, and what one participant contributes increases what another participant can understand or accomplish. One contribution can enrich the next contribution, which can spark another valuable interaction, and the value begins to compound across the network.

I have started thinking about this as the Intelligence Network Effect.

And that brings me back to ASSET 3.0:

A: Atomic Unit of Value
S: Synthesize the Unit of Value
S: Spark Demand
E: Exponential Value
T: Transform Intelligence

A lot is packed into those five steps, especially once we start asking what the Atomic Unit becomes when humans and agents participate together, how demand works when agents can create demand for other capabilities, and what happens when the resulting intelligence becomes a proprietary asset for the enterprise.

I will unpack all of that in my next post.

For now, the evolution I keep coming back to is much simpler. The internet gave us networks where human participation created proprietary data. Generative AI accelerated how humans could create and consume inside those networks. Agentic AI introduces participants that can reason, decide, act and contribute their experience back into the network.

We are moving from data network effects toward intelligence network effects.

And the network has a new participant.