According to an internal memo obtained by Business Insider, Meta's Superintelligence Labs is trying to speed up AI development by guiding teams to new tools that allow developers to build and test products more quickly.
This shift highlights meta CEO Mark Zuckerberg's interest in large-scale bets on AI. Over the past year, Meta has poured hundreds of millions of dollars into seducing top researchers with a nine-digit wage package and reorganising AI work under the new META Superintelligence Labs (MSL). Now, the new leader is revamping the internal engineering tools that those teams use. This is a shift that emphasizes the urgency of the meta moving towards keeping up with rivals such as Openai and Google.
Internal documents show that MSL's products and Applied Research (PAR) Group, led by former Github CEO Nat Friedman, are stoking staff from Meta's slow, homemade infrastructure towards mainstream developer platforms such as Vercel.
One note from late September shows that Meta's existing systems, designed for billions of users and a huge engineering team, are “too long” to deploy changes and “doesn't encourage coding,” making it difficult for small and medium-sized AI teams to experiment with. Vibe coding practices, or writing codes for AI tools, are becoming more widely popular among engineers in the high-tech industry. Meta, on the other hand, has built an internal alternative called “nesting,” which more easily integrates with Meta's data for rapid experiments.
Another internal post shows that Meta pairs with Github, a widely used code hosting and collaboration platform owned by Microsoft, to provide a faster way to work for PAR teams.
Meta's new top AI executives, Friedman and Alexandre Wang, are Versel investors. Vercel has become one of the biggest names in the market for building web and AI apps. In September, it raised $300 million at a $9.3 billion valuation from backers such as Accel, Singapore's GIC, BlackRock, Khosla Ventures and General Catalyst. The company's customers include giants such as Netflix, Adobe and Stripe.
This memo offers a rare glimpse of how Meta is trying to accelerate AI ambitions by relying on external development and engineering tools and bypassing the limitations of its own systems built in different eras.
Meta declined to comment. Vercel and Github did not respond to requests for comment from Business Insider.
Related Stories
Business Insider tells the innovative stories you want to know
Business Insider tells the innovative stories you want to know
Metasystem “does not promote vibe coding”
In one note sent in late September, MSL's Head of Infrastructure Aparna Ramani told staff that the company's internal systems built to support billions of users and large developer teams are not suitable for small, fast engineering teams within PAR.
“The change takes too long (time) vs. (A few minutes), and the overall tech stack does not encourage coding atmospheric,” writes Ramani.
In other words, engineering tasks that can be deployed quickly and easily elsewhere often require more manual work in the meta and slow down the team.
Ramani's notes outline the plan with two approaches to quickly build app prototypes and reduce deployment time within 99 minutes.
The first approach uses Vercel. It writes that it is “widely used and can be unblocked from MSL immediately.” She writes that Vercel is willing to work with Meta to ease some data and hosting restrictions. Ramani also writes that Meta already has one tool running in Vercel and there are several other projects in development.
The second approach is to build an internal platform called “Nest” in Meta's system to host apps built with TypeScript, a programming language written by Microsoft, widely used throughout the industry. Ramani writes that a working prototype is expected within two weeks of the memo.
Nest ultimately becomes a “de facto” option, and keeping Vercel available provides an “escaping valve with a functional gap,” the memo said.
Meta uses Vercel and Github
The second internal memo, published on September 17th, shows how Meta uses Vercel on GitHub to provide a faster way to provide prototypes to the PAR team. Meta already runs 10 projects on Vercel and Github, which now updates “in a few minutes”. The post said.
Meta often looks to external technology when it lacks its own tools. Earlier this year, Business Insider reported that after employees said Meta's own Llama AI model was struggling with complex tasks, Meta deployed an internal coding assistant called DevMate, which uses rival AI models, including humanity's Claude.
Meta recently announced a feature called Vibes, which uses external AI image generator Midjourney, rather than its own image generation technology. Friedman is an advisor to the Mid Journey.
Any hints? Please contact Pranav Dixit by email to AT pranavdixit@protonmail.com Or at the signal 1-408-905-9124. Use personal email addresses, non-working WIFI networks, and non-processed devices. Here's a guide to sharing information safely.
