How Is AI Accelerating Meta's New App Launches?

Meta's Return to the Standalone Apps Arena

Meta has long sought to extend its digital reach beyond the boundaries of Facebook, Instagram, and WhatsApp, yet its previous attempts at building standalone social apps fell short of expectations. Today, the company is revisiting that path with fundamentally different tools: large language models that have become the backbone of its product development strategy.

What's Different This Time?

During its Q2 2025 earnings call, CEO Mark Zuckerberg revealed that Meta has recently launched a series of new apps, including:

• Seller: A standalone app aimed at Marketplace sellers.
• Forum: A dedicated app for Facebook Groups as a separate experience.
• Instagram Instants: A fast visual experience within the Instagram ecosystem.
• A photos app and an AI-powered bedtime stories experience.

What sets this wave apart from its predecessors is the reliance on large language models to shorten the development cycle and enable engineering teams to test ideas at a far greater pace than was previously possible.

Lessons Learned from Past Failures

This strategy cannot be understood without revisiting history. In the mid-2000s, Meta established an internal lab called Creative Labs, which produced several apps including Slingshot, Rooms, Paper, Moments, and Riff — all of which were shut down by 2015 without attracting a sufficient user base.

In the early 2010s, Meta tried again through an internal R&D team known as the NPE Team, which launched more than ten apps including Bump, Aux, BarS, Tuned, and others. None survived, and they were discontinued one after another.

This pattern reveals that the problem was never a shortage of ideas, but rather slow execution and the difficulty of acquiring early users in a fiercely competitive environment.

Threads: The Model Redrawing the Map

Threads stands as the clearest real-world proof of this new approach's success. The app has reached 500 million monthly active users, and Zuckerberg is counting on it to become one of the company's billion-user apps in the future.

Threads' growth rested on two key pillars: first, leveraging Instagram's existing user base to drive large numbers immediately at launch; and second, deploying large language models within its recommendation system — something company executives described as yielding "tangible gains" in the quality of suggested content and increased user engagement.

The Role of Large Language Models in Product Architecture

CFO Susan Li explained to investors how these models operate on two complementary levels:

• Enhancing existing systems: The models understand the true context of content and generate more precise training data, improving the efficiency of ranking and recommendation algorithms.
• Assisting engineers: AI agents evaluate content quality, detect emerging trends, and test ranking changes before they go live.

Li also noted that Meta reached a significant technical milestone earlier this year: every Reel and every post in the Instagram feed is now automatically analyzed by a large language model that identifies its topic and tone, which in turn improves the accuracy of recommendations delivered to users.

What Can Users Expect Soon?

Zuckerberg indicated that new consumer products are in the pipeline and will be launched in the near future, without disclosing further details. Meta appears to be betting on a new formula: the development speed that language models provide, combined with the sheer power of its billions-strong user base, to transform small ideas into scalable products at an unprecedented pace.

Will this formula succeed in breaking the pattern of failure that has accompanied its previous ventures? The answer hinges on the market tests ahead.

✦ بقلم فريق دروب أيديا