Bluesky leans into AI with Attie, an app for building custom feeds

Here is the rewritten article.

---

Bluesky’s Bet on Attie: Can AI Save Decentralized Social Media from Itself?

For years, the promise of a decentralized social network has felt like a ghost in the machine—always teased, never quite materializing. Bluesky, the brainchild of former Twitter CEO Jack Dorsey, has spent its formative years building the plumbing: the At Protocol, a decentralized framework designed to wrest control of social data away from corporate overlords. But plumbing, as any engineer knows, does not make a home. It requires a compelling faucet, a reason to turn the water on. That faucet, it turns out, is artificial intelligence.

On a quiet Tuesday, Bluesky Social PBC launched Attie, a new application currently in beta that leverages AI to let users build highly customized feeds [1]. This is not merely a feature update; it is a strategic pivot. After a period of relative silence focused on protocol development, Bluesky is finally shipping a user-facing product that attempts to solve the single greatest friction point of decentralized networks: the lack of a good, default experience. Attie is the company’s attempt to prove that you can have algorithmic curation without the algorithmic tyranny.

The Alchemy of User-Defined Rules

The core premise of Attie is deceptively simple: replace the black-box algorithm with a user-defined rulebook. Instead of a platform deciding what you see based on opaque engagement metrics, Attie allows users to define filters and prioritization rules [1]. Think of it as a personal news editor that never sleeps, but one that you program yourself. You tell it to prioritize posts from specific communities, filter out certain keywords, or surface content based on temporal or engagement thresholds.

This is where the AI comes in. While Bluesky has remained tight-lipped about the specific models powering Attie [1], the engineering challenge is immense. A rule-based system alone is brittle; it requires the user to anticipate every edge case. An AI-powered system, however, can infer intent. It can learn that when you say "show me tech news," you actually mean "show me news about open-source LLMs and vector databases," not "show me the latest smartphone unboxing." This semantic understanding is the secret sauce that turns a rigid filter into a fluid discovery engine.

The technical architecture likely relies on a hybrid approach. The At Protocol provides the decentralized data layer—the raw feed of posts. Attie then sits on top, acting as a curation layer. It uses AI to parse the semantic meaning of posts, classify them, and then apply the user’s rules [1]. This is a significant departure from the "one size fits all" recommendation engines that dominate platforms like X (formerly Twitter) or Facebook. It represents a move toward a more agentic web, where the user, not the platform, is the primary stakeholder in the content equation.

The Suno Effect: Why Customization is the New Black

Bluesky’s timing is impeccable. The launch of Attie comes at a moment when the tech industry is collectively realizing that users are tired of being passengers. They want to drive. This sentiment is perfectly encapsulated by the recent release of Suno v5.5, an AI music generation platform that introduced features like "Voices," "My Taste," and "Custom Models" [2]. Suno’s release notes explicitly state that "Voices" was the most requested feature [2], underscoring a universal truth: users want AI to work for them, not the other way around.

The parallel between Suno and Attie is striking. Suno allows you to shape the sonic texture of a generated song; Attie allows you to shape the informational texture of your social feed [1]. Both are responses to a growing fatigue with generic outputs. In the world of AI, "one model to rule them all" is giving way to a more modular, customizable paradigm.

This trend is not limited to consumer apps. Consider the case of Intercom, which developed Fin Apex 1.0, a specialized customer service AI model. The results were staggering: Fin Apex 1.0 achieved a 73.1% resolution rate, outperforming general-purpose giants like GPT-5.4 and Claude Sonnet 4.6, which managed only 71.1% [3]. Intercom invested a staggering $100 million in this development [3]. The lesson is clear: specialization beats generalization in specific contexts. Attie, by focusing exclusively on the context of social feed curation, is following the same playbook. It is a specialized model for a specialized task, designed to outperform the generic "For You" page [1].

The Developer Dilemma: Opportunity vs. Centralization

For the developer community, Attie represents a double-edged sword. On one hand, the app’s reliance on the At Protocol creates a rich environment for innovation. Developers who understand both decentralized architecture and AI deployment are suddenly in high demand [1]. The ability to build custom feed algorithms that run on a decentralized backbone is a powerful proposition. It opens the door for niche communities—say, a feed for alpine weather enthusiasts or a feed for AI researchers—to curate their own digital spaces with surgical precision.

However, the complexity of this integration cannot be overstated. Integrating AI features like Attie’s custom feeds requires deep expertise in both protocol architecture and AI model deployment [1]. This creates a high barrier to entry. While the At Protocol aims for decentralization, the AI layer could inadvertently concentrate development efforts around Bluesky’s core team [1]. If Attie’s AI models remain proprietary and closed, the promise of a truly open ecosystem is compromised. The risk is that Bluesky becomes the "Apple of decentralized social"—a beautiful, curated walled garden built on open land.

This is the central tension of the project. To succeed, Attie needs to attract users. To attract users, it needs a polished experience. To get a polished experience, Bluesky may need to control the AI stack tightly [1]. The question is whether they can eventually open that stack up to third-party developers without sacrificing quality. The success of platforms like WordPress (which offers both a hosted solution and a self-hosted one) suggests it is possible, but it requires a deliberate, long-term investment in developer tooling and open-source models.

The Business of Attention: Disrupting the Algorithmic Oligopoly

From a business perspective, Attie is a direct assault on the advertising-driven revenue models of incumbent platforms. For the last decade, social media has been a game of "attention extraction." Platforms optimized for time-on-site, using opaque algorithms to keep users scrolling. Attie flips this script. By giving users control over their feed, Bluesky is betting that quality of engagement will trump quantity of engagement [1].

If users embrace customizable feeds, the dominance of algorithmic curation erodes [1]. This could enable new business models. Instead of selling user attention to advertisers, Bluesky could sell premium AI processing power—allowing users to run more complex, computationally expensive feed rules. Or, it could license the Attie AI stack to enterprises looking to build internal social networks. The cost of developing and maintaining this AI infrastructure is a key variable. While Bluesky has not disclosed its AI costs [1], the Intercom example suggests a substantial investment is required [3]. However, the potential for increased user engagement and data control could justify these costs [1].

The winners and losers in this new ecosystem are becoming clearer. Developers specializing in decentralized AI and At Protocol integration stand to benefit [1]. Platforms that rely on opaque data practices and algorithmic manipulation may face user attrition [1]. The losers will be those who cannot adapt to a world where the user, not the platform, holds the keys to the feed.

The Bigger Picture: A Reckoning for the At Protocol

The mainstream narrative around Bluesky’s Attie launch often focuses on the technical AI integration [1]. But the critical strategic implication is being overlooked: the potential impact on the decentralized social media movement itself. While the At Protocol provides the infrastructure for decentralized networking, Attie offers a tangible incentive for user adoption [1]. It is the killer app that the protocol needed.

However, there is a profound risk. By building a powerful, AI-driven front-end, Bluesky risks inadvertently centralizing control over the At Protocol [1]. If Attie’s models become proprietary and inaccessible to third-party developers, it could undermine the very decentralization principles the protocol was built upon [1]. The reliance on AI, while offering immediate benefits, introduces a new point of vulnerability and potential centralization [1].

This is the paradox of the modern AI stack. To be useful, AI models need data and compute. To be decentralized, protocols need to distribute control. Reconciling these two forces is the defining challenge of the next decade of social media. Bluesky’s Attie is the first major test case. Can you have a powerful, AI-curated social experience that is still genuinely owned by its users? The answer is not yet clear, but the experiment has begun.

---

References

[1] Editorial_board — Original article — https://techcrunch.com/2026/03/28/bluesky-leans-into-ai-with-attie-an-app-for-building-custom-feeds/

[2] The Verge — Suno leans into customization with v5.5 — https://www.theverge.com/entertainment/903056/suno-ai-music-v5-5-model

[3] VentureBeat — Intercom's new post-trained Fin Apex 1.0 beats GPT-5.4 and Claude Sonnet 4.6 at customer service resolutions — https://venturebeat.com/technology/intercoms-new-post-trained-fin-apex-1-0-beats-gpt-5-4-and-claude-sonnet-4-6

[4] MIT Tech Review — The Download: the internet’s best weather app, and why people freeze their brains — https://www.technologyreview.com/2026/03/27/1134755/the-download-best-weather-forecasting-app-why-people-freeze-brains/