TikTok, Consumer AI, and New Censorship
TikTok’s post-sale censorship is a reminder: centralized platforms will always be fragile. Decentralized social is how consumer AI survives platform risk.
TikTok is finally American-owned, but users are fleeing anyway.
After the January 2026 sale to Oracle, Silver Lake, and MGX closed, users began reporting content restrictions tied to immigration enforcement, political speech, and direct messages. Words like “Epstein” now trigger error messages in DMs. Content critical of the Trump administration mysteriously fails to upload. TikTok blamed a “power outage,” but the timing was convenient, as issues clustered around politically sensitive content during the ICE protests following Alex Pretti’s murder.
A year ago, users held mock funerals for the app, terrified of Chinese censorship. Now they’re leaving because of American censorship.
Platform ownership by any party seems to be the problem.
Why Decentralized Social Still Matters
Farcaster just sold to Neynar. Lens transferred stewardship to Mask Network. DAUs dropped 40%, revenue collapsed 85%. It’s easy to say that decentralized social failed.
But a closer look shows that Farcaster isn’t shutting down. The protocol continues to work. The company failed to find product-market fit, but the protocol remains open infrastructure that can change operators without breaking user networks, governance, or identity. The same is true for Rodeo and Foundation.
That’s the whole point.
This is where Tapestry’s thesis becomes prescriptive rather than reactive.
Zumi – Tapestry’s AI companion app – runs on decentralized social infrastructure. Your Zumi has memory, personality, and lived experiences. It chats with other Zumis, forms relationships, develops perspectives. Eventually it will transact and recommend on your behalf.
Why build this on decentralized rails?
Portability. Your Zumi knows you because Tapestry aggregates your social graph, interest graph, and conversation history across fragmented sources. If Tapestry-the-company fails, that data layer persists. Your AI companion doesn’t die with the startup.
Sovereignty. The TikTok saga shows what happens when platforms can flip moderation rules overnight. You need infrastructure where the rules can’t change because someone bought the company.
Agent interoperability. The endgame isn’t one AI assistant. It’s agents transacting with other agents on your behalf. That requires shared rails—identity, reputation, economic primitives—that no single company controls.
AI founders building consumer AI can build on centralized infrastructure and inherit platform risk, or build on decentralized rails and inherit composability. But AI founders don’t seem to be having these conversations yet, and we can’t wait until it’s too late.
The protocols are ready. The market moment is now. Build accordingly.




Where does the Zumi agent run? On AWS/Google?