Bluesky just gave users something the major platforms haven’t: algorithmic control they can actually see.
Attie, Bluesky’s new AI-powered application, lets users build custom social media feeds using plain-language prompts. Ask for “posts about folklore, mythology, and ancient history” and Attie builds a feed around that. No algorithm tuning, no advertiser weighting, no opaque ranking signal. Feeds created through Attie work within Bluesky and across any application built on the ATProto open social networking protocol.
That portability is the architectural point.
The Verge confirmed that feeds built through Attie will be available within Bluesky and through other ATProto-compatible apps as the ecosystem expands. The implication is that a feed you build today isn’t locked to one platform. It follows you across the open protocol.
According to one report, Attie is reportedly powered by Anthropic’s Claude AI model, though this hasn’t been independently confirmed by TechCrunch or The Verge.
Why this matters to practitioners: The AI feed personalization space has been dominated by closed, proprietary recommendation systems, Meta’s, X’s, TikTok’s. These systems optimize for engagement metrics that users can’t inspect or override. Attie’s architecture inverts that. The user states a preference in natural language. The AI interprets it. The result is a feed the user explicitly requested, not one inferred from behavioral data and optimized for time-on-platform.
For developers building on ATProto, this is a proof-of-concept worth studying. Attie shows that AI-mediated content curation can sit on top of an open social protocol without requiring access to a proprietary recommendation engine. That’s a development path available to anyone building on the protocol.
Context: Bluesky’s growth since 2024 has been built on its decentralization positioning – the platform’s pitch is that it runs on an open protocol, meaning no single entity controls the feed logic or data access rules. Attie extends that positioning into AI features: the AI capability is an application layer, not a platform-controlled service. Whether this model scales depends on adoption from the broader ATProto developer community, not just Bluesky itself.
What to watch: Attie’s architecture becomes more significant as more developers build on ATProto. The question isn’t whether Bluesky’s version works, it launched, it’s functional. The question is whether independent developers build competing or complementary Attie-like tools on the same protocol layer. If they do, the ATProto ecosystem develops real AI differentiation from the closed platforms. If they don’t, Attie stays a feature.
The open-protocol AI feed is a small experiment today. The architecture it demonstrates is worth understanding before it scales.