Meta is developing an AI model router named Switchboard through its internal AI incubator, AAI Labs, aimed at reducing costs by automatically assigning simpler coding tasks to less expensive models. This project, which is still in its early stages and may not be released as a product, fits within a broader initiative that encompasses about 200 approved projects in AAI Labs focused on developer tools and internal infrastructure. Switchboard functions similarly to OpenRouter’s Auto Router by scoring request difficulty and routing tasks accordingly.
Meta: Meta is a major technology company operating social media platforms and advancing AI capabilities across its products and infrastructure. Through its internal AI incubator AAI Labs, Meta is developing Switchboard to optimize AI model selection for coding tasks and reduce associated expenses. The project reflects Meta’s focus on internal efficiency tools that could also scale to other organizations using AI coding agents.
AAI Labs: AAI Labs serves as Meta’s internal AI incubator responsible for exploring new projects in consumer products, developer tools, and infrastructure. It is currently advancing Switchboard as one of its approved initiatives to handle AI model routing for cost savings on coding workloads. The lab operates with a portfolio approach to test and potentially deploy tools that support Meta’s broader AI efforts.
Switchboard: Switchboard is an early-stage AI model router under development at Meta’s AAI Labs that evaluates task difficulty and automatically directs simpler coding requests to lower-cost models. It draws inspiration from existing routing solutions and is intended to help cut internal AI expenses, with possible future use by external teams running large-scale AI coding agents. The project remains experimental and may stay limited to internal applications.
Project Scope: Switchboard is part of a broader set of initiatives inside AAI Labs focused on developer tools and internal infrastructure.
AI Model Routing: Switchboard evaluates request difficulty to route coding tasks between models, mirroring the approach of OpenRouter’s Auto Router.
