Bluesky’s latest experiment turns feed design into a chat prompt
Bluesky’s team has introduced a new standalone app called Attie, an AI assistant designed to let people create custom social feeds using natural-language instructions instead of code. The project, unveiled at the Atmosphere conference by former Bluesky chief executive Jay Graber and chief technology officer Paul Frazee, is built on top of the AT Protocol, the open social networking framework underlying Bluesky itself.
The idea is straightforward but significant. Rather than manually assembling a feed through filters, lists, or custom ranking logic, a user can simply describe what they want to see. In the example provided with the launch, someone could ask for posts about folklore, mythology, and traditional music, with an emphasis on Celtic traditions. Attie then translates that request into a working custom feed.
That makes Attie less a conventional recommendation engine and more an interface for building one. Bluesky has already made feed choice a central part of its product identity, distinguishing itself from platforms where a single algorithm dominates what users see. Attie extends that logic by aiming to lower the barrier to participation. If it works as intended, users would no longer need technical knowledge to shape the information layers sitting on top of a social graph.
Why this matters for open social platforms
The launch also says something broader about the direction of open protocols. Bluesky’s argument has long been that social networking should be built more like the web, with portable identities, interoperable services, and room for multiple applications to compete on user experience. Attie suggests the company believes AI can accelerate that model by making customization easier for non-developers.
Graber framed the shift in explicitly democratic terms. In a blog post cited alongside the announcement, she said the AT Protocol was built so anyone could create applications on top of it, but in practice that still mostly meant people who could code. Agentic coding tools, she argued, change that equation by making it increasingly possible to personalize software without programming experience.
That is an important distinction. Many consumer AI products are being positioned as assistants inside closed ecosystems. Attie is being presented instead as a way to manipulate an open data layer with a defined schema. In other words, the AI is not just helping users consume content. It is helping them assemble the logic that governs how content is organized and surfaced.
Today it builds feeds, later it may build apps
For now, Attie’s practical scope is limited. Custom feeds made with the assistant are initially confined to the standalone app, and the service is in closed beta with a waiting list. But Bluesky says the plan is to make those feeds available inside Bluesky and other apps built on the AT Protocol.
That near-term roadmap would already be meaningful. Feed portability across multiple AT Protocol apps could give users a stronger sense that their preferences belong to them rather than to any single platform. It could also encourage developers to build more niche or specialized applications, knowing that discovery tools and personalization layers may become easier to create and share.
The longer-term ambition is broader. According to the announcement, users will eventually be able to use Attie to “vibe code” entire applications on top of the protocol. That phrase carries a lot of hype, but the underlying point is clear: Bluesky sees AI-assisted software creation as a way to widen participation in building social tools.
If that vision materializes, the company would be pushing beyond feed customization into a model where AI becomes a bridge between user intent and application development. The implication is that open protocols could become more useful not just because they are open, but because AI lowers the cost of actually doing something with that openness.
Attie is an early test of AI-native social infrastructure
There are still plenty of unanswered questions. The announcement does not detail how accurate or controllable these AI-generated feeds are, how moderation issues will be handled, or what limits will exist when users begin generating more ambitious software behaviors. Closed beta status also means the public cannot yet judge how reliably the system converts vague prompts into durable, high-quality feed logic.
Even so, Attie stands out because it is not simply another chatbot bolted onto a consumer app. It is a test of whether generative AI can become an interface layer for open network infrastructure. That is a more structural use case than summarizing posts or drafting replies. It suggests a world where people describe the online experience they want and software assembles it on demand.
For Bluesky, that is strategically coherent. The company has spent much of its existence arguing that users should have more control over identity, moderation, and ranking. Attie is an attempt to make that control more accessible. Instead of choosing among tools built by others, users may eventually be able to describe their own.
Whether that leads to better social experiences will depend on execution. But the launch makes clear that Bluesky is trying to align two major technology trends at once: open social infrastructure and AI-assisted creation. If Attie succeeds, it could make personalization on decentralized platforms feel less like configuration and more like conversation.
This article is based on reporting by The Verge. Read the original article.
