Meta uses LLMs to speed app launches and recommendation testing
Meta is using large language models to launch standalone apps faster and says more consumer products are coming soon. During the second-quarter earnings call, CEO Mark Zuckerberg cited Instagram Instants, Forum for Facebook Groups and Seller for Marketplace. He also pointed to Threads, now at 500 million monthly active users.
AI speeds up product experiments
Zuckerberg said AI is speeding product development and should make it easier to ship more apps. Meta plans to build out additional ideas and use recommendation systems to scale them. It also says AI is making its core apps more relevant and improving results for businesses.
Meta has pursued standalone social apps before. Creative Labs produced Slingshot, Rooms, Paper, Moments and Riff, but the effort ended in 2015 after the apps failed to find an audience. The later NPE Team tested Bump, Aux, Move, Spark, CatchUp and Hotline; none became a breakout success, and they were shut down.
How LLMs improve recommendations
CFO Susan Li described two ways LLMs are improving ranking. They help existing systems understand what content is about and generate better training data. LLM-powered agents also assist engineering work by evaluating content quality, detecting trends and testing ranking changes.
Meta reached another milestone earlier this year: every Reel and Feed post on Instagram is now automatically processed through an LLM for analysis of topic and tone. The company is also building LLM-native recommendation systems that could help new apps scale.
Threads offers a working example
Threads is Meta's current example of AI helping an app scale. The company seeded it from its existing user base and promoted it across Facebook and Instagram. Meta also reports “significant gains” from AI-powered content recommendations, while Zuckerberg says Threads could eventually become a billion-user app.
Meta did not identify the next apps; investors focused instead on AI spending and enterprise ambitions. Zuckerberg said the new consumer products are “releasing soon.” For businesses considering the same approach, the practical point is to pair faster testing with distribution and a clear check on whether each product finds an audience.