OpenRouter's case study of Descript distinguishes changing the model from changing the company hosting it. Descript, an AI video and audio editing company, originally wrote fallback logic that switched to a different model when a request failed. Through OpenRouter, it now configures alternative inferenceRunning a trained model to get answers — the phase where AI is actually used, as opposed to trained.Full definition → providers (the companies that host a model) for the same model, keeping model selection constant when a host becomes unavailable. Descript also routes only to providers within its data-retention policy. This separates a reliability decision, where to run a request, from a product decision, which model should answer.
Terms in this piece · Glossary
inference — Running a trained model to get answers — the phase where AI is actually used, as opposed to trained.
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
eval — A repeatable test for AI quality — a set of tasks plus scoring — used the way software teams use test suites, because model output is too variable to judge by eyeballing.
