Bluesky has stepped into the AI space with Attie, a standalone application that lets users build custom social feeds through natural-language prompts rather than code. The app was presented at the Atmosphere conference, where attendees became the first beta testers. By tying AI-powered feed creation to the open AT Protocol, Bluesky is making a bet that decentralized social tools can compete with the algorithmic engines of major platforms. But the early product raises questions about how much control users will actually gain.
What Attie Does and How It Works
Attie is designed to let people describe the kind of feed they want in plain English and then have the app generate it automatically. A user could, for example, ask for a feed of posts from trusted journalists covering a specific beat, and Attie would assemble that stream using data available through the AT Protocol. Jay Graber, who outlined the project on her newsletter, frames it as an “agentic social app and custom feed builder” with the goal of making building on the AT Protocol accessible to people who have no programming background.
The app runs on Anthropic’s Claude as its underlying AI model, according to reporting from TechCrunch. Users sign in through an AT Protocol or Atmosphere login, which ties Attie directly to the same identity layer that powers Bluesky itself. That design choice means Attie is not a walled-off experiment. It sits within the same decentralized ecosystem, and any feed it creates could, in theory, be shared or remixed by others on the protocol.
Graber has called Attie the “first agentic social app on atproto,” a label that signals ambition beyond simple feed curation. The word “agentic” implies the AI does not just filter content but actively acts on a user’s behalf, selecting, sorting, and assembling posts based on ongoing instructions. Whether that distinction holds up in practice will depend on how much autonomy the app exercises and how transparent its decisions are to the people using it.
Who Built It and Who Runs Bluesky Now
Bluesky interim CEO Toni Schneider described Attie as a standalone product built by Jay Graber. That framing is notable. Rather than folding AI features into the main Bluesky app, the company chose to ship a separate application. This keeps the core social network uncluttered while giving Graber room to experiment and iterate without destabilizing the main user experience.
According to Graber’s own description, she currently holds the title of Bluesky Chief Innovation Officer, a role that appears oriented toward building new tools on top of the protocol rather than managing day-to-day operations of the social network. Positioning Attie under that innovation umbrella reinforces the idea that it is both a testbed for new ideas and a signal to third-party developers that Bluesky wants experimentation to happen at the app layer, not just inside the flagship client.
The decision to position Attie as its own product, rather than a feature update, also carries strategic weight. Standalone apps can attract different user bases, fail independently without dragging down the parent brand, and iterate faster. If Attie gains traction, it could become a proof of concept for developers who want to build their own AI-driven tools on the AT Protocol. If it stalls, Bluesky’s main app stays untouched, and the company can treat the experience as a learning exercise about how users actually want to direct AI agents in social contexts.
Beta Access: Open Signups, Closed Doors
The current state of Attie’s availability carries some ambiguity. Graber’s Liquid Frontier post states that beta signups are open, inviting people to express interest in trying the app. At the same time, a post on Bluesky describes the beta as invite-only and closed. Both statements can coexist if signups are open but access is granted selectively, though neither source spells out that distinction clearly.
What is confirmed is that the initial beta testers are attendees of the Atmosphere conference, giving the earliest access to a self-selected group of protocol enthusiasts and developers. This rollout strategy mirrors how Bluesky itself launched, with invite codes creating scarcity and buzz before wider availability. The approach builds anticipation but also limits early feedback to a narrow, technically literate audience. Whether Attie works well for casual users who simply want a better feed will not be clear until the beta expands beyond conference attendees and into the broader Bluesky user base.
There is also a practical question about expectations. If people can sign up but most will not receive access right away, Attie risks creating frustration among exactly the curious, early-adopter crowd that might otherwise help refine the product. Balancing the need for a controlled test environment with the desire to showcase progress publicly will be one of the first tests of how Bluesky manages AI-driven experiments under its brand.
AI That Serves Users or AI That Replaces Judgment
The pitch behind Attie rests on a simple premise: people should be able to shape their own social feeds without relying on opaque algorithms controlled by platform operators. Graber’s Liquid Frontier post frames the goal as making AT Protocol tools accessible through natural-language descriptions, removing the technical barrier that currently limits who can build on decentralized networks and who can meaningfully customize what they see.
That vision sounds appealing, but it also introduces a tension that current coverage has only hinted at. When a user tells Attie to build a feed of “trusted journalists,” the AI has to decide what “trusted” means. Those decisions are not neutral. They reflect the training data, the model’s biases, and whatever ranking logic Attie applies behind the scenes. Swapping one black box (a platform algorithm) for another (an AI agent interpreting vague prompts) does not automatically give users more control. It changes who builds the box.
This is not a hypothetical concern. Anthropic’s Claude, which powers Attie, is a capable language model, but like all large language models, it can produce inconsistent or unpredictable outputs depending on how prompts are phrased. A feed request for “balanced political news” could yield very different results than one for “independent political analysis,” even if the user intends the same thing. The quality of Attie’s output will depend heavily on how well it translates casual language into precise filtering rules, and on whether users can inspect and adjust those rules after the fact.
Transparency will be key. If Attie merely presents finished feeds without exposing why particular accounts or posts are included, users may feel just as disempowered as they do under conventional engagement-optimized algorithms. On the other hand, if the app surfaces its reasoning (showing which criteria it inferred from a prompt and letting people tweak them), it could meaningfully increase user agency. The difference between an “agent” and a recommendation engine may come down to how reversible and legible its choices are.
What This Means for Decentralized Social Tools
Attie represents a clear bet that AI can lower the barrier to entry for decentralized social media. The AT Protocol has always been technically open, meaning anyone can build on it, but “anyone” in practice has meant developers comfortable with APIs and data schemas. If Attie works as described, it could turn non-technical users into feed builders, expanding who gets to shape the network’s information flows.
For Bluesky, that shift could be strategically important. A decentralized protocol is only as vibrant as the ecosystem around it. If AI-powered tools like Attie make it easier to create niche feeds (local news digests, hobby-specific streams, or curated lists of experts on narrow topics), the result could be a richer, more diverse landscape of experiences than a single, centrally controlled timeline can offer. In that sense, Attie is not just a convenience feature; it is a potential catalyst for the protocol’s growth.
At the same time, Attie highlights unresolved questions about governance and responsibility in decentralized systems. If third-party agents are effectively deciding what users see, who is accountable when those feeds amplify misinformation, harassment, or low-quality content? The AT Protocol’s openness means anyone can build an agent, but Bluesky’s endorsement of Attie as a flagship experiment will likely shape norms around safety expectations, disclosure of AI involvement, and user recourse when things go wrong.
Ultimately, Attie is an experiment in whether AI can make decentralization feel less like a developer playground and more like a user-friendly alternative to mainstream social platforms. Its success will depend less on the novelty of natural-language prompts and more on the depth of control it actually hands over, including how clearly it explains its choices, how easily people can override them, and how well it integrates with the broader ecosystem of AT Protocol apps and feeds.
For now, Attie sits at an intriguing intersection: a decentralized protocol, a powerful language model, and a promise that ordinary users can direct both. Whether that promise holds up under real-world use will shape not only Bluesky’s AI strategy but also the broader conversation about what “user agency” means in an era where algorithms increasingly mediate every social interaction.
More from Morning Overview
*This article was researched with the help of AI, with human editors creating the final content.
