Networks
Where the fediverse gets its footing
After a decade of near-misses, decentralised social software finally started to feel — for a few thousand people — normal. What did the software actually do differently?
The most surprising thing about the current wave of decentralised social software is not that it works. Decentralised software has worked, technically, for a long time. The surprising thing is that ordinary people — non-technical, protocol-illiterate, largely uninterested in the politics of it — are using it as if it were ordinary software. That did not happen the last three times.
What changed is not the protocol. The protocol is roughly what it was five years ago, with better tooling around it. What changed is that a small number of clients decided to hide the seams. Server selection was made invisible, or nearly so. Follow relationships across servers were made to look like follow relationships within them. Moderation tools were built to work across the fabric of the network, not just on one node.
This is a familiar shape. Every successful decentralised technology — email, git, the phone system — went through a period in which the interface stopped exposing the underlying topology. You do not think about which mail server you are on. You do not think about which git remote you are pushing to, until you have to. The fediverse has just started to allow the same forgetting.
The word for what these clients did is abstraction, but the honest word is concealment. They did not remove the servers, the relays, the moderation boundaries; they hid them behind an interface that behaves as if they weren't there. This is a design choice with teeth. The moment concealment fails — a server goes down, an admin turns hostile — the whole topology reappears at once, and the newcomer who was promised simplicity discovers they were standing on a machine the entire time.
Whether that is a betrayal or a kindness depends on how often the concealment holds. Email's abstraction holds well enough that most people will never learn what an MX record is, and that ignorance costs them almost nothing. The open question for federated social software is whether it can reach that same standard — whether the seams can be hidden not just prettily but durably, so the forgetting is safe and not merely convenient.
The tradeoff is that the seams still exist and they still matter. A server admin can, in a bad afternoon, make a lot of your day worse. A server can go quiet and take your archive with it. Federated moderation is fragile in a specific way that centralised moderation is not. These are not solved problems. They are ongoing negotiations, and the current wave has just made them feel more like weather than like earthquakes.
It helps that the people running the servers this time seem to understand what they signed up for. The better admins talk about their instances the way a good landlord talks about a building — as an ongoing responsibility to actual people, not a growth asset to flip. That posture is not encoded in any protocol. It is a culture, and cultures are more fragile and more valuable than code.
The failure mode I worry about is not collapse but consolidation: a slow drift back toward two or three enormous servers that everyone joins by default, at which point the network is federated in theory and centralised in practice. The topology would still be open. The power would not be. Decentralisation is a property you have to keep choosing, not one you achieve once and bank.
What I take from this is a small optimism about the medium. Not that decentralised social media will overtake the centralised kind — it probably won't — but that a serious alternative now exists, at a size that supports weird communities. That is enough. It doesn't have to be everything. It just has to be somewhere.