Bluesky Attie: Build Your Own Social Media Algorithm With AI | Cliptics

What if you could actually tell your social feed what to show you?

Not in some vague "I'm interested in tech" way that platforms ignore. I mean really design it. Like talking to a friend who curates your feed exactly how you want it.

That's the idea behind Attie, Bluesky's new AI app they just unveiled at their Atmosphere conference this March. And honestly, the more I learn about it, the more I wonder if this is how social media should have worked all along.

The Announcement That Caught My Attention

Jay Graber, Bluesky's former CEO and now chief innovation officer, introduced Attie alongside CTO Paul Frazee at the Atmosphere conference late March 2026. She said something that stuck with me: "We think AI should serve people, not platforms. An open protocol puts this power directly in users' hands."

That philosophy shows up in how Attie works. It's not part of the main Bluesky app. It's a separate thing built on top of the AT Protocol, the open framework Bluesky runs on. Which means anyone could potentially build something similar if they wanted to.

The conference attendees became the first beta testers. They got to try building custom feeds just by typing commands in plain English, the same way you'd chat with any AI. And under the hood? It's powered by Anthropic's Claude.

So you've got this combination: an open protocol, a powerful AI model, and natural language processing. That's what makes Attie different from the usual "adjust your preferences" settings buried in app menus.

How It Actually Works

Here's what I find fascinating about the whole setup.

You don't need to understand how algorithms work. You don't need to know what "engagement weighting" or "recency decay" means. You just describe what you want in conversational language.

Something like: "Show me tech news but skip anything about crypto drama. I want photography posts from people I follow, and surface interesting discussions even if they're a few days old."

And Attie translates that into an actual custom feed configuration. It's leveraging Claude's ability to understand context and intent, then mapping that to the technical specifications the AT Protocol needs.

What's really interesting to me is that this isn't just filtering. It's feed construction. You're not removing stuff from an existing algorithm. You're building your own from scratch.

That raises questions I keep turning over. How specific can you get? Can you combine multiple criteria in complex ways? What happens when your natural language description is ambiguous? Does it ask for clarification or make assumptions?

I don't have all those answers yet since it's brand new. But the concept itself feels like a genuine shift.

Why This Matters More Than It Seems

Most social platforms give you the algorithm they want you to have. Sure, you can mute words or hide certain content. But the fundamental logic controlling what you see? That's locked down. It's optimized for engagement, ad views, whatever metrics the platform prioritizes.

Bluesky's taking a different approach with Attie. They're saying the algorithm should work for you, not the other way around.

And because it's built on the open AT Protocol, the entire process is more transparent. You're not dealing with a black box that mysteriously decides what's "relevant." You're creating explicit rules that you can understand and modify.

That transparency matters. When you know why something appears in your feed, you can make better decisions about how to adjust it. You're in control instead of just reacting to what the algorithm throws at you.

But here's where it gets complicated. About 125,000 users have already blocked Attie's Bluesky account, making it the second most blocked account on the network. That's a lot of people saying "no thanks" right out of the gate.

I've been wondering why. Is it AI skepticism? Concerns about how the tool might be used? A general discomfort with algorithmic feeds period? Or maybe people just don't want to do the work of designing their own experience when passive scrolling is easier?

Whatever the reason, it shows that even when you give people control, adoption isn't guaranteed. Changing how we interact with social media takes more than just better technology.

The Bigger Picture I Keep Thinking About

This experiment with Attie connects to larger questions about how AI fits into our digital lives.

We're seeing AI integrated into everything right now. Search engines, productivity tools, creative software, and now social media curation. But most of that integration happens behind closed doors. Companies decide how to use AI, and we just live with their choices.

Attie flips that. It puts AI in your hands as a tool to shape your own experience. Instead of AI optimizing for platform goals, it's optimizing for your stated preferences.

That distinction feels important to me. Not because Attie is perfect or because everyone will use it. But because it demonstrates an alternative model. One where AI serves individual users directly rather than primarily serving the platform's business model.

And since it's built on open protocols, other developers can build similar tools. Or different tools. Or completely unexpected applications we haven't imagined yet. That's the advantage of open systems over closed ones.

What I'm Curious About Next

I keep asking myself questions I don't have answers to yet.

How will people actually use this in practice? Will they create highly personalized feeds that fragment the social experience even more? Or will interesting feed designs emerge that people share with each other?

What happens when your carefully designed algorithm still doesn't feel quite right? Do you keep tweaking it forever, or do you eventually just accept some level of imperfection?

And what about the learning curve? Natural language processing is powerful, but it's not magic. You still need to think clearly about what you want. That might be harder than it sounds when you're dealing with something as complex and emotional as social media.

There's also the AI content creation angle. If everyone's feed is customized differently, how does that change content strategy? Do creators need to think differently about how their posts get discovered?

Where This All Might Go

I don't know if Attie specifically will take off. The early blocking numbers suggest a rocky start. And people's social media habits are deeply ingrained. Changing them takes time and compelling reasons.

But the underlying idea resonates with me. The notion that we should be able to design our own information environment rather than accepting what's handed to us. That algorithms should be tools we control, not forces that control us.

Bluesky's approach with the AT Protocol and now Attie represents one possible future. An open one where interoperability and user control matter more than lock-in and engagement optimization.

Whether that future actually arrives depends on a lot of factors. Technical challenges, user adoption, competitive pressure from existing platforms, regulatory changes, and probably things we can't even predict yet.

But experiments like this show us what's possible. They expand the range of what we think social media could be. And sometimes that's valuable even if the specific implementation doesn't succeed.

I'm going to keep watching how this develops. What people build with it. How Bluesky iterates on the concept. Whether other platforms respond with their own versions or double down on their existing approaches.

Because at the end of the day, this isn't really about Attie or Bluesky or even social media specifically. It's about who gets to decide how technology shapes our experience. And whether we're willing to do the work of shaping it ourselves.