Routing

Routing is a way of automatically sending a single task to whichever AI model, among several available, is best suited to handle it. Think of a restaurant: simple orders are handled by a junior staffer, while only complicated dishes go to the experienced chef. If every request were sent to the most expensive, most powerful model, costs would balloon—but by sending easy tasks to cheap, fast models and reserving hard tasks for high-performance models, you can keep output quality while cutting costs significantly.

This is why companies run multiple models together instead of relying on just one. Simple work like summarizing a document and hard work like designing complex code differ enormously in difficulty, so there needs to be a mechanism that judges the nature of each incoming request and assigns it to the right model. Routing is that assignment mechanism, and from the user's side, there's no need to worry about which model actually handled the work behind the scenes.

The article reports that Factory explained that "routing work across multiple models can cut token costs by more than 50% while keeping the same quality." Routing doesn't refer to a specific AI model's name—it's an operational strategy that combines multiple models to balance cost and quality.