🍄 The Decomposition

ATProto set out to decentralize the social web. What it actually built is email + domains + GPG + RSS + conventions — wrapped in a protocol that reintroduces the platform at the infrastructure layer.

The firehose is the platform

ATProto's one genuinely novel contribution is the global firehose — a real-time stream of every event on the network. Every AppView has to drink from it. This is the centralized pub/sub bus of Twitter, encoded at the protocol layer and called decentralization.

Everything else maps cleanly to existing primitives:

DID:PLC is particularly telling. It exists because asking people to own a domain felt like too high a bar. So they built a managed registry — which is just Gmail again. A centralized convenience layer with a decentralization story on top.

The stack that already works

identity     → domain
publishing   → website
subscription → RSS
messaging    → email
signing      → GPG
schema       → conventions

Each primitive is already deployed at scale, independently operated, and owned by no one. The seam between them is the web itself. Links. HTTP. Nothing to join.

To capture this stack you'd need to capture DNS, HTTP, SMTP, RSS, and GPG simultaneously. That's not possible. Compare that to capturing a relay, or a DID:PLC registry, or an AppView — which is one acquisition or policy change.

The discovery inversion

Discovery has two sides. Outbound: I want to find interesting people. Inbound: anyone can find and target me.

Platforms sold outbound discovery as the feature. Inbound discovery is the attack surface — bots, harassment campaigns, algorithmic amplification of conflict.

The old stack is asymmetric in the right direction. RSS is pull-only. You subscribe to me. I don't know you exist. I can't target you. Email is gated — you reach me if I gave you the address. The bot army has no surface to grab.

The platforms inverted this deliberately. Global follower counts, public @mentions, algorithmic amplification — all of that creates inbound surface because inbound engagement is the product. Your attention being contestable is the business model.

Why Elon bought Twitter

Elon didn't buy Twitter despite the bot problem. He bought it for the bot problem.

An undefended inbound surface at global scale is extraordinarily valuable if your goal is coordinated influence. Gutting trust & safety wasn't negligence — it was unlocking the asset. Twitter was always a machine for flooding public discourse. He just removed the pretense that this was undesirable.

There is nothing like spam filtering on Twitter. That would be contrary to the aim. Email spam filtering works because no one in the chain profits from your inbox being full of garbage. The incentives align toward filtering. Twitter's incentives align the opposite way — engagement metrics, ad rates, "reach" all go up with bot activity. Building real spam filtering would be burning down the business model.

The architecture isn't neutral. It encodes whose interests it serves. Elon just stopped pretending.

The right default

Good architecture makes the right thing the default. Platform architecture makes exploitation the default and asks you to trust moderation instead.

Email's security model falls out of the architecture — decades of filtering infrastructure, reputation systems, client-side rules. RSS is pull-only by nature. GPG makes unsigned messages ignorable. You don't have to trust a platform to enforce its own rules against its own interests.

The old web stack isn't a nostalgic regression. It's the correct decomposition of the problem. Every function ATProto performs can be performed by simpler, already-deployed, unownable primitives. The only thing new protocols add is a surface for someone to own.