Consumer agents will have network effects

In consumer agents are the next big thing, I wrote about how consumer AI would move up the stack, from capturing the 3% of consumer spend on digital goods, to consumer agents which take the other 97%.

Now, Instinct, Grok Bot, and Muse are here, and consumer agents are officially the current big thing. I have been a user of Instinct since mid-July, about two weeks after I wrote the prior post, and consumer agents have become a part of my daily ritual.

So these are clearly incredible products, but are they good businesses? What staying power do they have?

Perhaps most importantly, for the history of consumer technology: do they have network effects?

This article is going to argue that the answer is yes.

To briefly credentialize myself, I worked at Uber from 2015-17 as a launcher and driver growth lead, and have invested in a number of network effects and marketplaces businesses with

Creator Ventures since (as well as co-authoring a Harvard Business School academic paper on network effects in 2018 which according to Claude has been independently cited a total of two (2) times).

So I have at the very least been thinking about network effects for over a decade.

So without further ado, here's why I believe that consumer agents do, in fact, have network effects.

The definition of a network effect, according to HBS, is that the value of a product increases with the number of users who leverage it. For consumer agents, edge-cases are a big part of this.

Various X users have shared "nightmare" Instinct outcomes: it unintentionally cancelling flights, getting banned from Resy, or unintentionally leaking information. Clearly, consumer agents are not yet 100% reliable.

But these scenarios should be interpreted positively: they are proof of the network effects. This is a market where once you have built the product, the objective is to eliminate each 0.1% edge-case bug, one after the other. You need a lot of users to find them.

I wanted to do some research to support my sizable personal investment in Figma stock, so I spun up Instinct to go to the Wayback Machine, and find how regularly SaaS businesses were updating their landing pages. It took around an hour, but eventually it came back with a breakdown of which companies had made material changes to their landing pages, and how that's changed over time.

(The answer was that these changes had increased, especially between 2024-early 2026, during which coding agents progressed significantly).

This task took Instinct several hours, to browse, collect and analyze the materiality of landing page changes on the Wayback Machine.

It gave me the answer, accurately.

But the important point here is that now it knows.

If a future user asks the same question, it will be able to use some part of the work it did for me, for the next user. At least to do it faster (e.g. it could remember which of the landing page updates it looked at were material changes, and which were not).

At Google, where 15% of searches per day were historically unique and had never been searched before (the percentage is likely significantly higher now because of AI), consumer agents can use pooled intelligence to improve answers of subtly different queries.

Put simply, would you rather use a consumer agent that has been asked billions of questions, browsed trillions of sites, and spun up quadrillions of sub-agents? Yes, because it will have a much better view of the state of the world.

Multi-sided marketplaces are often a sufficient, but not necessary condition for network effects. Plenty of companies and products, from Google search to Waze, have exceptionally strong network effects, without having a marketplace or two sides.

The other argument against consumer agents having network effects is that there is a limited pool of compute, and by using it up, more users actually degrade the quality of the service (it's highly likely that consumer agents will start struggling under the strain of the next 100 million DAUs). However, I'd argue that, like Uber, (compute/driver) supply constraint is a capacity problem, not a supply constraint.

So yes, Instinct, Grok Bot and Muse have network effects. There will be one, or a small number, of winners in this market. In this case, the Silicon Valley excitement about the category is warranted.

And the best personal agent will probably turn out to be the one everyone else is using.