This page was translated from Dutch with AI.

Let your heavy model plan and your workhorse execute

In a training session about Claude I got a question I love to hear. I had just had a plan made and then I set the model one level lower. 'Why are you putting that on Sonnet now? Because you used Opus before that?' That was indeed the intention. And I notice that many people skip that step. They choose one model and let it do the entire task, from the first idea to the final file.

Perhaps you recognise that. You turn on the heaviest model, because you want things to go well. And then you also let it do the boring execution. That works, only you pay the full price for every step. Your limit is therefore used up faster than you would like.

Copy the prompt at the bottom of this article. With it you let a heavy model write a plan that your workhorse can execute afterwards.

Cartoon
Laat je zware model plannen en je werkpaard uitvoeren

Two phases in one task

Almost every larger task actually has two parts. First there has to be some thinking about what exactly you want, what can go wrong and in what order you do it. After that it just has to happen. Moving files, writing texts, building a structure.

For that first part a heavy model is appropriate. In Claude that is Opus, the model for strategic thinking and planning. In Copilot you get close with the Think deeper mode. For the second part you do not need that. Sonnet is actually your workhorse. If you already have a briefing and a plan, Sonnet executes that perfectly. In the training session I put it like this: plan in Opus, execute in Sonnet.

A heavy model is by the way at its best when it is not allowed to do anything yet. I literally add: 'Do not do anything yet, but formulate a plan.' That way it gets the space to think, without rushing headlong into the work.

Frequently asked questions

Why should I switch models within one task?

Planning requires different work than executing. A heavy model thinks deeper about what you want and what can go wrong. A workhorse executes a clear plan just as well and consumes less of your limit.

Do I have to open a new chat for the execution?

I usually recommend that. Save the plan as a file and point to it in a new session. Then your workhorse starts with a clean context and only what it needs.

Does this also work outside of Claude?

The principle works in every tool with multiple models or thinking modes. In Copilot you can for example plan with Think deeper and then continue working in the regular mode.

Where can I learn this hands-on?

In Claude AI Power User and in AI agents you work yourself with plans, briefings and different models, under guidance. You can find the dates at Workshops & Events.

Keep reading for free

Register free with your e-mail address and read on. Your account opens the free articles in the learning space; the rest is for subscribers, from Learn Plus up.

Already registered? No account yet?

No comments yetRead the comments

Comments

Members comment here and read each other's comments. Register for free or sign in to join the conversation.

Register for free or sign in

No comments yet. What do you take away?

    QR code of this page

    coachsteff.live/en/plannen-met-opus-uitvoeren-met-sonnet

    Scan with your phone to open the page.