Matching Model to Task

Task routing

Part of: Choosing the Right Model

A hospital triage nurse doesn't send every patient to the trauma surgeon. A sniffle gets a quick check; a car crash gets the full team. Good AI systems do the same thing, they triage tasks to tiers , sending easy work to cheap fast models and hard work to the heavy hitters. What it is Task routing (or model routing) means picking the model tier based on how hard the task is, rather than using one tier for everything. The two ends of the spectrum: - Simple work → small tier. Classification (spam / not spam), extraction (pull the date, pull the email), short formatting, yes/no decisions. Clear inputs, narrow outputs. - Hard work → large tier. Multi-step reasoning, long synthesis, ambiguous open-ended writing, anything where a wrong early step poisons the rest. Most apps land somewhere in the middle with a few clearly-hard and clearly-easy tasks at the edges. Route the edges, default the middle. How it works A router is just a function from "task" to "tier." It can be as simple as a rule table: Notice the router doesn't call any model, it just decides which one to call . That separation keeps the logic testable and the cost predictable. Why it matters Routing is where the cost-and-qua

Challenge: Routing Engine