Everyone is building LLM routers, we deprecated ours
Summary
The Manifest blog argues against model routing, sharing that routing to multiple LLMs rarely cuts costs and adds complexity. After real-world usage across thousands of users, they deprecated their own LLM router, advocating sticking with a single battle-tested model and relying on caching to reduce costs. The piece emphasizes reliability, predictability, and observability as higher priorities than marginal cost savings.