🧠 Stop burning your Claude usage on the wrong thing

Do this instead

🧠 Stop burning your Claude usage on the wrong thing

Came across a short today with a workflow that actually holds up, so I checked it.

The core idea: separate thinking from doing.

🎯 Use the strongest models (Opus 5, Fable 5) only for planning and strategy.

⚡ Hand the grunt work to cheaper models (Sonnet 5) running as subagents.

🔁 Optional: add /goal and Claude keeps working across turns until your condition is met, even after you walk away.

💛 The part I like as a psychologist: this technique is cognitive load management, applied to a machine.

🌱 Reserve the expensive attention for decisions that matter, delegate the rest.

👀 Curious: is anyone here already running a setup like this?

Originally published on Substack.