On network effects

Most products do not have network effects. The vast majority grow by addition: each new customer adds value at a steady rate, and the line goes up roughly in proportion to spend. You hire more, you ship more, and the line goes up at the same slope. Most companies are this, and there is nothing wrong with addition. Addition produces revenue.

A small number of products grow by multiplication. Each user makes the product more valuable for the next, and at some inflection the curve bends sharply upward. The work doesn't get harder. It compounds in a way that resists being copied — not because the work is special, but because the asset accumulating underneath is a different shape. Multiplication produces a place.

But the phrase is overused. People will tell you a product has network effects when it has a chat feature, a referral program, or any surface where users meet. Most of these are not network effects. They are nice features. The test of a real network effect is whether the experience changes with density. If a user joining late benefits from everyone who came before, you have one. If their experience is roughly the same regardless of who else is there, you have a fine product.

The cold start is the only thing that matters

Every network looks like a failure for a long time. Below a certain density it does not work. You carry the cost of seeding it without yet enjoying any of its returns. The math is unforgiving: a graph with three nodes and one edge has the same shape as a graph with three thousand nodes and one edge — which is to say, no shape at all.

The companies that survive this period do one of two things. They give the product enough single-player utility that the first user can show up alone and still get something useful, which is what keeps them around long enough for someone else to arrive. Or they assemble the first dense pocket by hand, on purpose, in one room or one school or one block, and let that one pocket prove the model before anyone tries to extend it.

The mistake is to skip past either of those — to assume the curve will bend simply because the pitch deck says it will. It almost never does. Most products that claim network effects die in the cold start.

The substrate is what makes the effect work

The early consumer internet got the strongest network effects of the last century because the cost of being a producer was real. The domain, the server, the HTML, the time to learn — these did the work an explicit identity layer would otherwise have had to do. The networks built on top of that substrate inherited the filtering for free. Most of what got called a network effect was downstream of the entry tax, not a property of the product itself.

When the entry tax goes to zero, the effects downstream of it stop working too. AI is taking the entry tax to zero. Anyone can spin up an agent, generate a video, post a thousand tweets in a minute. The networks built on the old substrate are losing the silent layer that was doing half their work.

The next generation of networks will have to re-import that layer explicitly. The thing that filters now will not be the cost of being on the network. It will be the cost of being verified to be on it.

The geometry from here

A managed identity layer is the precondition for network effects that survive contact with infinite supply. The networks that matter from here will be the ones that decided, on purpose, to make joining expensive again — not in dollars, but in commitment, accountability, and verifiable connection to a real person. That is not a step backward toward gatekeeping. It is the only configuration in which trust scales at the same rate as participation, and therefore the only configuration where the line bends upward instead of flattening.

The rest will look like networks for a quarter or two and then quietly stop compounding. Their curves will resume the slope of addition. The metric will be the same. The geometry will be different.

Most of the work, in the meantime, is patient. Planting in the right soil. Watering long after anyone is watching. Trusting that geometry — eventually, if you have the right substrate — does its own work.