[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"$fHujT8wmv0QBKGyaKyc5Q2nCCODNWqrOhXO6x3_38foA":3},{"article":4,"related":18},{"id":5,"slug":6,"title":7,"seo_title":8,"description":9,"keywords":10,"content":11,"category":12,"image_url":13,"source_guid":14,"published_at":15,"created_at":16,"updated_at":17},629,"blueskys-attie-the-ai-powered-feed-customizer-redefining-social-media","Bluesky Attie Turns Algorithm Choice From Ideology Into Product","Bluesky's AI Feed Customizer: How Attie Works","Bluesky's new Attie tool uses Claude AI to let users build custom feeds with plain English commands. See how it compares to traditional algorithm controls.","[\"Bluesky Attie\",\"AI feed customization\",\"AT Protocol\",\"custom algorithms\",\"social media algorithms\",\"Bluesky vs Instagram\",\"algorithmic choice\",\"feed generators\"]","\u003Cp>Bluesky just shipped the most consequential feature in its short history, and it has nothing to do with growing its user count. Attie, an AI assistant powered by Anthropic's Claude that lets anyone build a custom feed algorithm using plain English, transforms algorithmic choice from an ideological talking point into an actual product. Unveiled at the Atmosphere conference in late March 2026 by co-founder Jay Graber (now chief innovation officer) and CTO Paul Frazee, Attie represents something the social media industry has discussed for years but never truly delivered: genuine user control over the recommendation engine. The timing is not accidental. It arrives during a leadership transition, a competitive squeeze, and a broader reckoning over what AI should actually do inside social platforms.\u003C\u002Fp>\u003Ch2>The Three-Year Arc From Protest Network to Product Company\u003C\u002Fh2>\u003Cp>To understand why Attie matters, you need to trace Bluesky's evolution from a Twitter skunkworks project to an independent company navigating an identity crisis.\u003C\u002Fp>\u003Cp>When Bluesky launched its public beta in early 2023, its core pitch was decentralization via the AT Protocol. Custom feed generators existed from almost the beginning. Developers could spin up their own algorithm servers, declare them via a record in their AT Protocol repo, and let users subscribe. The feed generator starter kit on GitHub made this possible for anyone comfortable with TypeScript and server deployment. In theory, this was revolutionary. In practice, fewer than 2% of users ever touched a custom feed beyond the defaults.\u003C\u002Fp>\u003Cp>The problem was never philosophical. It was mechanical. Building a feed generator required standing up a server, writing code to index the Bluesky firehose, implementing ranking logic, and maintaining uptime. Even technically proficient users rarely bothered. The result was a two-tier experience: a tiny developer class that shaped discourse through custom feeds, and a vast majority of users passively consuming whatever Bluesky's default algorithm served them. The decentralized promise rang hollow when the reality looked identical to every other platform.\u003C\u002Fp>\u003Cp>Bluesky grew anyway, nearly 60% in 2025, climbing from roughly 26 million to over 41 million registered users. But growth decelerated hard. The platform went from adding 5 million users per month at peak to approximately 1.4 million per month by mid-2025. Daily active users hovered around 3.5 million as of late 2025, a roughly 12.8% DAU-to-MAU ratio that signals a retention problem more than a discovery problem. Bluesky needed a reason for casual users to stay, not just a reason for Twitter refugees to sign up.\u003C\u002Fp>\u003Cp>Enter Attie. By wrapping feed generation in a conversational AI interface, Bluesky collapses the entire complexity stack. You type \"show me posts about electronic music from people I follow\" and Attie writes the feed logic, deploys it, and surfaces it in your home interface. No servers. No code. No understanding of how the AT Protocol works under the hood. This is not a minor UX improvement. It is a fundamental restructuring of who gets to participate in algorithmic governance.\u003C\u002Fp>\u003Ch2>Instagram Gave Users a Dashboard. Bluesky Gave Them a Compiler.\u003C\u002Fh2>\u003Cp>Bluesky is not the only platform responding to the algorithmic transparency movement. Instagram rolled out its \"Your Algorithm\" feature to all English-speaking users in early 2026, letting people see what topics Meta's AI thinks they like and manually adjust those preferences. On paper, this sounds similar to what Attie offers. In practice, the two approaches diverge so sharply they barely belong in the same category.\u003C\u002Fp>\u003Cp>Instagram's model is a preferences panel. You can see that Meta has categorized you as interested in \"fitness\" and \"travel photography,\" and you can toggle those interests up or down. But you cannot define novel categories. You cannot combine signals (\"show me fitness content only from accounts with fewer than 10,000 followers who post original research\"). You cannot create an entirely new ranking function. You are adjusting knobs on Meta's algorithm. Meta still decides what knobs exist.\u003C\u002Fp>\u003Cp>Attie is closer to giving users a compiler than a control panel. Because the AT Protocol treats feed generators as independent services with a standardized interface, Attie is not constrained by a fixed set of ranking signals. It can, in principle, generate feed logic that combines follower graphs, post content, engagement patterns, temporal signals, and user-defined semantic categories in any combination. The constraint is Claude's ability to interpret the request and generate valid feed code, not a predetermined menu of options.\u003C\u002Fp>\u003Cp>X, formerly Twitter, has taken a third path entirely. After xAI open-sourced its Grok-powered algorithm in January 2026, the technical details became public: a three-stage pipeline of candidate sourcing, neural network ranking, and heuristic filtering where conversation depth is heavily weighted. But the real story is the pay-to-play structure. Premium accounts receive roughly 10x more reach per post than free accounts, creating the largest algorithmic inequality of any major social platform. Users do not control the algorithm on X. They pay for preferential treatment within it.\u003C\u002Fp>\u003Cp>These three models represent genuinely different philosophies about power. Meta says: we will show you the levers but we decide what levers exist. X says: the algorithm is a marketplace and reach is a commodity. Bluesky says: here is an AI that writes algorithms to your specification. The third option is the most radical, and also the most fragile, because it depends on an AI system interpreting ambiguous human preferences correctly.\u003C\u002Fp>\u003Ch2>The Claude-Shaped Elephant in the Room\u003C\u002Fh2>\u003Cp>Attie runs on Anthropic's Claude. This is simultaneously the smartest and the riskiest architectural decision Bluesky has made.\u003C\u002Fp>\u003Cp>Smart, because Claude's instruction-following capabilities are well-suited to translating natural language into structured feed logic. The AT Protocol's feed generator spec is relatively constrained: a service receives a request with a user DID and pagination cursor, and returns a ranked list of post URIs. The code generation task is bounded enough that a frontier language model can handle it reliably. Bluesky gets a world-class AI capability without building an ML team from scratch.\u003C\u002Fp>\u003Cp>Risky, because Bluesky has now introduced a critical dependency on a single external vendor for its flagship new product. Anthropic's API pricing, rate limits, model availability, and content policies all become constraints on Attie's functionality. If Anthropic decides certain types of feed curation requests are problematic, or if Claude's safety training causes it to refuse edge-case but legitimate feed specifications, Bluesky inherits those limitations with no recourse.\u003C\u002Fp>\u003Cp>There is also a cost structure question that nobody in the launch coverage has addressed. Running Claude inference for every feed creation and modification request across millions of users is not cheap. Bluesky's recent $100 million funding round provides runway, but AI inference costs scale with usage in a way that traditional software does not. If Attie succeeds and millions of users start iterating on custom feeds, the compute bill could become a meaningful fraction of Bluesky's burn rate. The company will eventually need to decide whether Attie is a loss leader that drives retention, a premium feature that justifies a subscription, or an open specification that third parties can implement with their own model backends.\u003C\u002Fp>\u003Cp>That last option is the most interesting and the most aligned with Bluesky's decentralist ethos. Because the AT Protocol already supports pluggable feed generators, there is nothing preventing someone from building an Attie-equivalent that runs on an open-source model like Llama or Mistral. Bluesky could, in theory, define a standard \"AI feed assistant\" interface and let the market compete on implementation quality. Whether the company will actually do this, or whether Attie will become a proprietary moat, will reveal a lot about how seriously Bluesky takes its own decentralization principles.\u003C\u002Fp>\u003Ch2>Who Wins, Who Loses, and What Breaks\u003C\u002Fh2>\u003Cp>The immediate winners are Bluesky's existing power users, the people who understood the value of custom feeds but lacked the technical skills to build them. Journalists, researchers, community organizers, and niche hobbyists now have a tool that matches their information needs to a personalized algorithm. This is genuinely new. No other platform offers this capability at this level of expressiveness.\u003C\u002Fp>\u003Cp>The immediate losers are the small ecosystem of developers who built feed generator businesses and tools on top of the AT Protocol. Attie commoditizes their work overnight. If a user can describe a feed in plain English and get a working algorithm in seconds, the value of a hand-crafted feed generator drops to near zero. Some will pivot to building more sophisticated tools on top of Attie's output, but many will find their niche eliminated.\u003C\u002Fp>\u003Cp>The more interesting second-order effects play out over months. If Attie drives meaningful improvements in Bluesky's retention metrics, particularly that anemic 12.8% DAU-to-MAU ratio, it validates a model where AI serves user intent rather than platform engagement metrics. This puts pressure on every other social platform to offer comparable controls. Instagram's preference toggles will look primitive. X's pay-for-reach model will look exploitative.\u003C\u002Fp>\u003Cp>But there is a darker possibility. Custom algorithms optimized for individual preferences can create filter bubbles far more airtight than anything a platform algorithm produces. When Meta's algorithm shows you polarizing content, you can blame Meta. When your own AI-generated algorithm shows you only confirming viewpoints, you have built your own epistemic prison. The discourse around algorithmic harm has assumed a platform-as-villain model. Attie introduces a world where users are the architects of their own information distortion.\u003C\u002Fp>\u003Cp>Bluesky's moderation team will face novel challenges. If a user instructs Attie to \"show me posts that criticize [specific ethnic group],\" the system must decide whether to fulfill the request. Claude's safety training provides a first line of defense, but the boundary between legitimate content filtering and harmful curation is blurry. A request like \"show me posts debating immigration policy\" is clearly fine. \"Show me posts opposing immigration\" is arguably fine. \"Show me posts attacking immigrants\" is clearly not. The gradient between these is where Attie's real policy challenges live.\u003C\u002Fp>\u003Ch2>What Builders Should Watch For\u003C\u002Fh2>\u003Cp>For founders building on the AT Protocol or adjacent social infrastructure, Attie signals three things worth acting on.\u003C\u002Fp>\u003Cp>First, the feed generator layer is now an AI-native surface. Any tool, service, or product that assumes users will manually configure their social media experience is obsolete. The interface for algorithmic preference is natural language, full stop. Build accordingly.\u003C\u002Fp>\u003Cp>Second, Bluesky's dependency on Claude creates an opening for competitors who offer model-agnostic feed generation. An open-source Attie alternative that lets users choose their AI backend, or runs entirely on-device for privacy, would be immediately compelling to Bluesky's privacy-conscious user base.\u003C\u002Fp>\u003Cp>Third, the combination of decentralized protocol plus AI-generated algorithms creates a new design space that does not exist on any other platform. You can build services that compose multiple AI-generated feeds, that let users share and remix feed algorithms, that create marketplace dynamics around feed quality. The AT Protocol's composability was always theoretically powerful. Attie makes it practically accessible.\u003C\u002Fp>\u003Cp>For Bluesky itself, the next twelve months will answer the question that has dogged the company since its founding: can a platform built on ideological differentiation (decentralization, user control, algorithmic choice) convert those principles into a product experience that retains mainstream users? Attie is the most credible attempt yet. It takes the AT Protocol's most distinctive capability, pluggable feed algorithms, and wraps it in the most intuitive interface possible, a conversation with an AI.\u003C\u002Fp>\u003Cp>The real test is not whether Attie works technically. It is whether ordinary users, the ones who never heard of the AT Protocol and do not care about decentralization, find that building their own algorithm makes Bluesky worth opening every day. If they do, Bluesky has found its product-market fit in a place no one expected: not in being the anti-Twitter, but in being the first social network where the algorithm genuinely works for you. If they do not, Attie becomes another impressive demo that failed to change behavior, and Bluesky's 3.5 million daily actives continue their slow plateau. The $100 million in the bank buys time. Attie needs to buy engagement.\u003C\u002Fp>\n\u003Cscript type=\"application\u002Fld+json\">{\"@context\":\"https:\u002F\u002Fschema.org\",\"@type\":\"NewsArticle\",\"headline\":\"Bluesky Attie: Why AI Feed Control Changes Social Media Forever\",\"description\":\"Bluesky's Attie uses Claude AI to let anyone build custom feeds in plain English. Here's why this matters more than Instagram's algorithm controls and what it means for the platform wars.\",\"datePublished\":\"2026-03-29T21:44:41.000Z\",\"dateModified\":\"2026-03-29T21:44:41.000Z\",\"wordCount\":1967,\"publisher\":{\"@type\":\"Organization\",\"name\":\"Seedwire\",\"url\":\"https:\u002F\u002Fseedwire.co\"}}\u003C\u002Fscript>\n\u003Cscript type=\"application\u002Fld+json\">{\"@context\":\"https:\u002F\u002Fschema.org\",\"@type\":\"BreadcrumbList\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\u002F\u002Fseedwire.co\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"News\",\"item\":\"https:\u002F\u002Fseedwire.co\u002Fnews\"},{\"@type\":\"ListItem\",\"position\":3,\"name\":\"Bluesky Attie: Why AI Feed Control Changes Social Media Forever\"}]}\u003C\u002Fscript>","AI & Machine Learning","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1774828901552-qpjowglrogs.webp","0ae5b4455fbffea93120b5020413b714d7edac80c53bc8a4c9172a3b61d56930","2026-03-29T21:44:41.000Z","2026-03-30T00:01:43.256Z","2026-05-20 04:02:22",[19,26,33,40],{"id":20,"slug":21,"title":22,"description":23,"category":12,"image_url":24,"published_at":25},1160,"nvidias-ai-agent-pcs-disrupt-cpu-market","Nvidia's AI Agent PCs Disrupt CPU Market","Nvidia partners with Microsoft, Dell, and HP to bring AI agents to the masses, potentially disrupting the $200B CPU market with easy, safe, and useful AI sol...","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1780372896898-m3py8qjssb.png","2026-06-01T21:35:00.000Z",{"id":27,"slug":28,"title":29,"description":30,"category":12,"image_url":31,"published_at":32},1159,"minimax-m3-revolutionizes-enterprise-ai-with-unprecedented-performance-and-affordability","MiniMax-M3 Revolutionizes Enterprise AI with Unprecedented Performance and Affordability","MiniMax-M3 delivers frontier AI performance with 1M token context and native multimodality. Rivals GPT-5.5 and Gemini 3.1 Pro at a fraction of the price.","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1780358478324-2nbfzx936oo.png","2026-06-01T16:10:05.000Z",{"id":34,"slug":35,"title":36,"description":37,"category":12,"image_url":38,"published_at":39},1156,"ai-agent-bottleneck-permissions-not-performance-hold-key-to-success","AI Agent Bottleneck: Permissions, Not Performance, Hold Key to Success","Enterprise AI agents face significant hurdles due to permissioning issues, rather than model performance. This article explores the technical and operational...","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1780200072608-785cnnl3x7d.png","2026-05-29T22:27:49.000Z",{"id":41,"slug":42,"title":43,"description":44,"category":12,"image_url":45,"published_at":46},1154,"memo-revolutionizes-llm-upgrades","MeMo Revolutionizes LLM Upgrades","MeMo's innovative memory model enables seamless LLM upgrades without retraining, transforming enterprise AI capabilities. Discover the technical implications...","https:\u002F\u002Fseedwire.co\u002Fapi\u002Fimages\u002Farticles\u002F1780113688089-flkdnur6fh.png","2026-05-29T19:28:17.000Z"]