Building Cost-Smart AI Agents with Sonnet, Opus, and Fable on Make

In this practical breakdown, Andrew tackles one of the most common questions Makers ask about AI automation: which Claude model should you actually use, and when? With pricing that ranges from cheap to seriously expensive, picking the wrong model for the job can quietly drain your token budget. 🔍 Inside the Build: * The Sonnet Baseline: Learn why Sonnet 5 is Andrew's go-to for most automations — simple text conversions, basic questions, arithmetic, and general knowledge tasks — all at a fraction of the cost of the bigger models. * Opus as Triage: See how Opus 5 shines as an orchestration layer, ingesting information, routing it to sub-agents or sub-modules, and re-digesting what comes back before returning an answer to the user. * When Fable Earns Its Price Tag: Andrew explains why Fable 5 is reserved for large, multi-step, complicated tasks — think 100-page PDF analysis or coordinating multiple sub-agents — and why you should think twice before defaulting to it. * Effort Levels Explained: A walkthrough of the low-to-max effort setting on each model, and how it independently controls token usage and cost regardless of which model you've picked. * Private Spaces for Teams: Andrew shows how Make's private spaces let team members experiment with their own connections and API keys without touching the shared credit pool. Andrew's approach: start cheap, scale up only when the results demand it. This video is a perfect blueprint for Makers building AI agents who want powerful automations without an unpredictable bill. 🛠️ Featured Stack: Claude Sonnet 5, Opus 5, and Fable 5 (via Make AI Agents) * Make AI Agents (Sub-agent orchestration) * Private Spaces (Team credit and connection management) * Reasoning Effort Settings (Token usage control) 💡 Pro-Tip from Andrew: "Always start on Sonnet at low effort, then work your way up." Andrew highlights that maxing out Sonnet's effort can often outperform Opus at its lowest setting — so the cheapest model is always worth testing to its limit before reaching for something more expensive.