The REAL Series · Part 2 of 2

This is the second part of a two-part series. Part 1, The REAL Human Moat in an AI World, introduced four sources of enduring value: Relationships, Embodied Experiences, Actual Things, and Lived Judgment. This part asks how those sources interact with network effects once agents start acting for participants.

A familiar claim in technology is that a network becomes more valuable as more people join it. More buyers attract more sellers. More colleagues make a collaboration tool more useful. More people on a messaging service make it easier to reach the people who matter.

What happens when the participants are represented by agents?

The thesis

My strategic thesis—a hypothesis still to be tested—is that agents can learn interfaces and move context faster than people can. That compresses three costs that have protected many networks: the cost of searching for alternatives, the cost of learning a new system, and the cost of switching.

A person who joins a new platform has to learn its interface, rebuild habits, find the right people, and decide whether to trust it. An agent may learn the system, transfer context, compare alternatives, and coordinate with other agents in a fraction of that time. If the thesis holds, networks become easier to enter and easier to leave, which may accelerate both their formation and their collapse.

Three kinds of “network”

The thesis applies differently depending on what kind of network a business actually has. Three cases are worth separating.

A genuine network effect. One more participant makes the product more valuable for existing participants. Figma is a useful example: a new teammate joins a shared file, contributes to it, and makes collaboration more valuable for everyone already there. The value comes from people working together, and the number of accounts is only a rough proxy for it.

No network effect. A standalone AI writing tool may be useful to millions of people. If each person's experience is isolated, and one person's use does nothing to improve the product for anyone else, the business has scale. Scale has its own economics, and those economics are separate from a network effect.

A fragile network effect. Participants do add value for one another, but the network's hold on them depends mainly on friction: search that is hard to do elsewhere, cumbersome switching, or data and relationships that users cannot easily move. This network is real, and it is the one most exposed if agents reduce those frictions.

How agents could change network economics

For a fragile network, an agent that searches across services, compares options, and carries preferences and context from one system to another weakens the platform's ability to rely on inertia.

For challengers, the same capabilities could help new networks form faster. Agents can find relevant participants, translate between systems, and coordinate activity that previously required people to discover one another and learn a new tool. A new entrant may reach useful density sooner.

These are implications of the thesis, and they remain to be tested. Many networks rest on scarce supply, trusted identity, reputation, regulation, physical infrastructure, or relationships that software connections cannot recreate. Agents can lower coordination costs; the underlying value still has to come from somewhere.

Recent commentary explores adjacent questions. Andrew Chen (October 2026) asks whether agents themselves will develop network effects. MIT Sloan (May 2026) argues that AI is reconfiguring the search and transaction costs platforms have traditionally managed. McKinsey (May 2026) examines how companies move from AI as table stakes to AI as a competitive moat. These are perspectives and analysis, and they point to the same management question: what value does the platform create when an agent is acting for the participant?

The fifth moat—and REAL+

I see five potential sources of a moat. The four REAL sources—Relationships, Embodied Experiences, Actual Things, and Lived Judgment—can each be a standalone moat. Network effects are the fifth source.

A network effect can be a moat on its own. In an agentic world, one that rests mainly on scale or switching friction may prove less durable, because agents can learn it, route around it, or help participants replace it.

A network effect becomes a durable, strong moat when it is combined with one or more REAL moats. I call that combination REAL+. The term names a structure of advantage; one REAL moat is enough to qualify, and a business does not need all four. The network amplifies the human value, and the REAL moat gives the network something meaningful to compound.

Three illustrations show the difference. A network of trusted professional relationships carries far more value than a database of the same names. A community built around shared, embodied experiences is harder to reproduce than a feed. A marketplace with trustworthy reputation and real fulfillment offers more than its matching software. In each case, every new participant can deepen value for the others, and that value is hard for an agent to transfer elsewhere.

The leadership test

User count is the easiest number to report and the least informative. Four questions say more about durability:

  • What gets better for existing participants when another joins?
  • Can an agent learn, transfer, or reproduce that value quickly?
  • Which REAL moat—if any—makes the network valuable in a way that is hard to recreate?
  • Does AI strengthen that source of value, or merely make the current interface more efficient?

Network effects remain a source of advantage. If the thesis holds, agentic AI will make them form and unravel faster, and the networks that last will be the ones built around something REAL.