---
title: "Designing Agents That Stay Fast and Affordable"
description: "Whitepaper: how agent architecture decides AI spend at scale, from context retrieval and model routing to caching, and the choices that cut cost per task."
canonical: "https://acme-jaret.duckdns.org/resources/guides/agent-cost-curve"
site: "Acme"
type: "Whitepaper"
---

# Whitepaper | Designing agents that stay fast and affordable as they grow

Model spend grows with every agent you ship. This paper walks through the architecture choices that decide where that money goes: how context is retrieved, how requests are routed between models, and how state is carried across long multi-step runs.

## Learn how to design agent systems that stay fast, accurate and affordable as usage grows

- Why well-ranked, permission-aware context cuts wasted reasoning and lowers the tokens spent per answer.
- How routing each step to the right-sized model gets more done for every dollar of compute.
- Which signals platform leaders should track to keep cost and quality in balance.