Introduction
The line between social media and scientific research has never been thinner. In early 2026, Bluesky took a bold step: its AI assistant Attie, originally designed to help users draft posts and manage feeds, evolved into a full-fledged open social research tool. The shift isn’t just a feature update — it represents a new paradigm in how researchers and hobbyists interact with public conversation data. At the heart of this transformation is a concept called vibe coding: using natural language to instruct AI agents to perform complex data tasks without writing a single line of traditional code.
For years, social media research meant wrestling with APIs, writing Python scripts, and cleaning messy JSON payloads. Attie changes that. Now, anyone can say “Show me how sentiment about remote work changed among UK academics this year” and get a structured analysis back — complete with CSV exports and visualizations. This article is a deep dive into how Attie works as a research tool, how vibe coding makes it accessible, and what practical steps you can take to leverage it for your own projects.
From Chatbot to Research Assistant
When Bluesky launched Attie in 2025, it was positioned as a friendly assistant for navigating the decentralized social network. Users could ask Attie to summarize threads, find trending topics, or suggest replies. But Bluesky’s engineering team — inspired by the open-source ethos of the AT Protocol — soon realized that Attie’s underlying language model could be repurposed for structured data extraction.
According to Bluesky’s official announcement in February 2026, Attie’s research mode allows users to access the full firehose of public posts (subject to rate limits) and run customizable queries. The key innovation is the natural language query compiler: Attie translates plain English requests into AT Protocol queries without exposing the user to technical details. This is vibe coding in action — you describe the outcome you want, and the AI handles the implementation.
What Makes Attie Different?
| Feature | Traditional API Research | Attie Vibe Coding |
|---|---|---|
| Skill required | Python, API keys, SQL | Natural language |
| Query flexibility | High (but time-consuming) | High (instant) |
| Iteration speed | Slow — each change requires re-coding | Fast — just rephrase |
| Reproduction | Scripts can be documented | “Vibe recipes” / chat logs |
Attie’s expansion is not just about convenience. It lowers the barrier for researchers from fields like sociology, political science, and public health who may lack coding expertise but deeply understand their research questions.
Understanding Vibe Coding in Social Research
The term vibe coding was popularized by AI researcher Andrej Karpathy in 2025. It describes the practice of letting an AI write code based on high-level instructions, where the human focuses on “the vibe” — intent, creativity, and direction — rather than syntax. In the context of social research, vibe coding means you can tell Attie:
“Find posts from January to June 2026 that mention ‘universal basic income’ and ‘California’. Group them by weekly frequency and show the top 10 most-liked posts each week.”
Attie then:
1. Constructs the appropriate AT Protocol queries.
2. Paginates through results, honoring rate limits.
3. Aggregates and analyzes the data.
4. Returns a structured table or chart.
You never see a line of Python, and you don’t need to worry about authentication — Attie handles context via your Bluesky account permissions.
A Concrete Example: Analyzing Public Sentiment on AI Regulation
Let’s walk through a realistic research scenario. Suppose you’re a policy analyst tracking public opinion on AI regulation in the EU.
Step 1: Define your query in natural language.
“Show me all public Bluesky posts from March 2026 that contain ‘EU AI Act’ or ‘AI regulation’. Exclude posts by bots (identified by high posting frequency). Give me a table with columns: date, post content, like count, and user’s stated location (if available).”
Step 2: Attie processes the request.
Behind the scenes, Attie uses its internal language model to:
- Identify the keywords: EU AI Act, AI regulation.
- Set time range: March 2026.
- Access Bluesky’s public firehose (Bluesky provides a sample of public posts via the AT Protocol’s com.atproto.sync.subscribeRepos endpoint).
- Apply a bot-detection heuristic: flag accounts that post more than 50 times per day.
- Extract location from user’s profile description or labels (many users voluntarily add location).
Step 3: Receive results.
After a few seconds, Attie returns a Markdown table (which you can export as CSV):
| Date | Post content | Likes | Location |
|---|---|---|---|
| 2026-03-05 | “The EU AI Act is too strict on open source.” | 142 | Berlin, Germany |
| 2026-03-07 | “We need global AI regulation, not just EU.” | 89 | Paris, France |
| … | … | … | … |
Step 4: Iterate with vibe coding.
Want to focus on negative sentiment? Just say: “Now filter to posts with angry emojis or words like ‘unfair’ or ‘dangerous’.” Attie pivots instantly.
Practical Tips for Using Attie as a Research Tool
1. Start with Broad Queries, Then Narrow
Vibe coding works best when you iteratively refine. Begin with a broad question like “What are people posting about climate change this month?” Examine the results, then ask Attie to filter by language, region, or sentiment. You can even ask Attie to suggest refinements: “What sub-topics appear most frequently in this dataset?”
2. Use “Vibe Recipes” for Reproducibility
Attie saves your entire conversation history. To share or reproduce a research finding, you can export the chat log as a “vibe recipe” — a plain-text document that anyone can paste into Attie to get the same result (given the same data availability). This is analogous to publishing a Python script. For example:
```markdown
Comments