Josh Shaked’s “Exploring the Frontier of Feeds” makes a compelling case. Unbundling the algorithm from the platform is the right direction. The vision of portable, user-controlled feeds redistributing attention and wealth is genuinely exciting.

But the economic model rests on an assumption that deserves scrutiny: that fragmenting feeds will fragment value.

It won’t. Here is why.

Attention follows power laws, not averages

Shaked imagines 1M feeds each earning $196K. The arithmetic works. The economics don’t. Every open marketplace we have observed — the App Store, the Chrome Web Store, podcast directories, YouTube channels — converges on the same shape: a power-law distribution where the top 0.1% captures most of the value and the long tail earns close to nothing.

Why? Because choice costs attention. When a user faces 1,000 feeds, they do not sample 1,000 feeds. They pick the three that already have reputation, social proof, or visibility. The Visa interchange analogy breaks down here: Visa transactions distribute roughly in proportion to spending. Attention does not distribute in proportion to supply. It concentrates.

Open feeds will not escape this gravity. They will reproduce it in a new venue.

The missing question: who pays for the view?

Shaked frames platforms as “routers” for third-party feeds. But routing requires infrastructure. On ATProto, the cost gradient is steep: running a PDS is lightweight, but operating an AppView — the layer that aggregates, indexes, and serves data across the network — is expensive. Feed generators sit on top of AppViews. Every custom feed adds computational load to that layer.

The article is silent on this. Who bears the cost of serving a million feeds? If the answer is “a few large AppView operators,” then we have recreated the concentration problem one layer deeper in the stack. The platform is no longer the bottleneck. The AppView is.

This is the question the open feed conversation has not yet addressed: decentralizing the algorithm means little if the infrastructure beneath it remains centralized by economic necessity.

From distribution sovereignty to perspective sovereignty

The real frontier may not be who controls the feed, but who controls the view. Distribution sovereignty — choosing your own algorithm — is a necessary step. But it is not sufficient. What matters next is perspective sovereignty: the ability to define not just what you see, but how you see it, without depending on a single expensive intermediary to assemble that view.

ATProto’s cost gradient — where PDSs are cheap but AppViews are expensive — creates a bottleneck that recurses the very centralization the protocol was designed to escape. Solving this requires rethinking the cost structure of the stack itself. Not just who curates, but who can afford to serve.

The open feed vision is worth building. But if we stop at the algorithm layer without solving the infrastructure layer beneath it, we risk celebrating a redistribution that never arrives.