Bluesky just made feeds programmable.
The decentralized Twitter alternative launched Attie, an AI-powered app that lets anyone build custom feeds without writing code. No Python scripts. No API documentation. Just describe what you want, and Attie generates the feed algorithm for you.
This is either brilliant or completely unhinged, depending on how much you trust AI to understand what “show me good posts” actually means.
What Attie Actually Does
Attie sits on top of Bluesky’s AT Protocol, which already lets users create custom feeds. The difference? Before Attie, you needed technical chops to build those feeds. Now you just talk to an AI assistant.
Want a feed of cat photos from accounts with under 1,000 followers? Tell Attie. Want posts about AI that exclude anything mentioning crypto? Attie handles it. The app translates natural language into feed algorithms, then publishes them to Bluesky where anyone can subscribe.
It’s democratizing algorithmic curation, which sounds great until you remember that most people can barely articulate what they want for dinner, let alone what makes a good social media algorithm.
The Timing Is Suspicious
Bluesky launching an AI tool right now feels calculated. The platform has been positioning itself as the anti-algorithmic alternative to X and Threads, where users control their own experience. Adding AI-powered feed creation keeps that promise while acknowledging that most people don’t want to learn to code.
But there’s tension here. Bluesky’s whole pitch is transparency and user control. AI systems are famously opaque. When Attie builds your feed, can you actually verify it’s doing what you asked? Or are you just trusting another black box, except this one has a friendly chat interface?
The Real Test: Does It Work?
I haven’t tested Attie yet because it just launched, but the concept raises obvious questions. How well does it handle ambiguous requests? What happens when your natural language description conflicts with itself? If you ask for “interesting posts about technology,” what does “interesting” mean to the AI?
These aren’t theoretical concerns. A recent Stanford study highlighted the dangers of AI chatbots giving personal advice because they confidently provide answers even when they shouldn’t. Feed curation isn’t life-or-death, but the same problem applies: AI systems are really good at sounding certain about subjective judgments.
Why This Matters Beyond Bluesky
Attie represents a broader trend: making algorithmic systems accessible to non-technical users through AI interfaces. We’re seeing this everywhere, from no-code app builders to AI-powered analytics tools.
The question is whether this accessibility comes at the cost of understanding. When you write code, you’re forced to think precisely about what you want. When you describe something to an AI, you can be vague, and the AI will fill in the gaps. Sometimes that’s helpful. Sometimes it means you get something that sort of works but doesn’t quite match your intent.
For social media feeds, the stakes are relatively low. You try it, see if you like it, adjust. But this pattern of “AI as interface layer between humans and complex systems” is going to show up in higher-stakes contexts. Attie is a test case.
The Verdict (So Far)
Bluesky’s move is smart. They’re solving a real problem: custom feeds are powerful but inaccessible to most users. Attie could change that.
But I’m skeptical about whether AI is the right solution. The appeal of Bluesky is supposed to be transparency and control. Adding an AI layer that interprets your intentions introduces new opacity. You’re trading one kind of algorithmic mystery for another.
Still, I’d rather have this than nothing. If Attie makes custom feeds accessible enough that people actually use them, that’s progress. Just don’t pretend you understand your algorithm better because you described it in English instead of code. You’re still trusting a system to interpret your words, and that system has its own biases and limitations.
Bluesky is betting that most users will accept that trade-off. They’re probably right.
đź•’ Published: