Reel Friends: Building Social Discovery that Scales to Billions

2026-05-13 · Source: Engineering at Meta · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Software Development & Engineering · Depth: Intermediate, quick

Summary

Meta's new "Friend Bubbles" feature for Facebook Reels, which highlights Reels watched and reacted to by friends, is the subject of a recent Meta Tech Podcast episode. Hosted by Pascal Hartig, the episode features software engineers Subasree and Joseph from the Facebook Reels team. They delve into the extensive engineering work required to bring this seemingly simple social discovery feature to life, scaling it to billions of users. Key discussion points include the evolution of the underlying machine learning model, distinct user behaviors observed between iOS and Android platforms, and a crucial, surprising discovery that ultimately made the entire feature functional. The podcast underscores how straightforward user-facing features often demand profound technical innovation.

Key takeaway

For software engineers developing user-facing features, recognize that seemingly simple social discovery tools like "Friend Bubbles" demand significant underlying machine learning and platform-specific engineering. Your team should anticipate deep dives into model evolution and distinct user behaviors across platforms (e.g., iOS vs. Android) to achieve scalability and functionality. Don't underestimate the complexity hidden beneath a straightforward UI.

Key insights

Simple social features often mask complex engineering challenges, especially at scale.

Topics

• Friend Bubbles
• Facebook Reels
• Machine Learning Models
• Social Discovery
• Mobile Engineering
• Scalability

Best for: Software Engineer, Machine Learning Engineer, AI Engineer

Related on AIssential

• Bond, a new social media platform, wants to use AI to help you kick your doomscrolling habit — Bond uses AI to transform user-posted memories into personalized real-world activity recommendations.
• How Coinbase scaled AI to 1,000+ engineers | Chintan Turakhia — Large engineering organizations can achieve significant AI adoption and velocity gains through hands-on leadership and strategic tool integration.
• Steam Recommender using similarity! (Undergraduate Student Project) [P] — Granular, vector-based game descriptions enhance recommendation transparency and discoverability beyond broad tags.
• Monthly "Is there a tool for..." Post — Centralized community forums effectively manage recurring tool recommendation queries in AI.
• When Do LLMs Generate Realistic Social Networks? A Multi-Dimensional Study of Culture, Language, Scale, and Method — LLM-generated social networks are significantly influenced by prompt design, culture, language, and model scale.
• The Sample Complexity of Multiclass and Sparse Contextual Bandits — Algorithms achieve optimal sample complexity for s-sparse contextual bandits, closing a significant gap in prior work.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Engineering at Meta.