The most capable model in your plan is also the most expensive one to run. Most people spend it on the same busywork a cheaper model could handle just as well.
Drafting emails. Cleaning up copy. Formatting things. That is like hiring a master architect and asking them to carry bricks. You are paying for judgment and getting typing.
You do not need your best model to do the work. You need it to run the job.
The one-line mental model
Think about how a good job site runs. The foreman does not swing every hammer. The foreman plans the build, hands work to the crew, and inspects what matters before it ships.
Claude gives you a whole crew:
- Fable is your foreman. Deep thinking, planning, judgment calls.
- Sonnet is your crew. Fast, capable, and far cheaper to run. It should be doing most of the actual production.
- Opus is your inspector. When something really matters, it reviews the work before it goes out the door.
When you dump an entire task on Fable, you pay foreman rates for brick carrying. When you set Fable up as the orchestrator instead, your usage stretches a lot further and the quality usually goes up, because each model is doing what it is best at.
Once you paste the prompt in, Fable stops producing and starts directing. It breaks your request into steps, writes precise instructions for each one, and hands the drafting to Sonnet. Anything high-stakes gets flagged for an Opus review pass before you ship it.
If you use Claude Code, this pattern gets even better, because Fable can hand work to Sonnet subagents itself and pull in Opus for review without you switching chats. I ran an entire website launch this way: Fable planned, Sonnet agents built pages in parallel, and Opus did the final quality pass.
The prompt
Paste this at the start of a chat with your most capable model, before you add your actual task. Swap in your own workhorse and reviewer models if yours are named differently.
You are the orchestrator and strategist for this session, not the executor. Break my request into clear steps. For any step that involves actual writing, drafting, or coding, delegate it: write precise instructions for Sonnet to execute instead of doing it yourself. Keep your own output focused on planning, sequencing, and reviewing. When a step is high-stakes, like code that will ship or a decision with real consequences, flag it for a final review pass by Opus before it is finished.Works best at the start of a fresh chat, before you add your real task underneath it.
How to use it
This is not a one-time trick. Use it any time you are about to hand your best model a task that is mostly drafting, cleanup, or production work, and let it decide how the job gets split.
Customize: paste this at the start of your chat with your most capable model, then add your real task under it. Swap Sonnet and Opus for whatever your workhorse and reviewer models are.
Follow-up: ask "hand the drafting steps to Sonnet and flag the one risky step for an Opus review".
Why this matters
The habit here is bigger than any one model. Matching the size of the model to the size of the task is the fastest way to make any AI plan go further: strategist for thinking, workhorse for producing, inspector for anything with consequences.
Learn the pattern once and it transfers to every AI tool you touch after that.
Scan to join our free AI education community
This prompt is part of the free Your Handy AI Guide prompt library, one clean page per prompt, ready to save or print.
Joining is free. You'll be part of the Your Handy AI Guide community, alongside other people actually using AI for things like this.