I Made Claude Opus 5.5 Plan and Sonnet 5.5 to Work

substack.com · newsletter

Explore this email design and adapt it to your own brand. Review the copy, links, and offer before sending.

Email content

Forwarded this email? Subscribe here for more

A practical system for using Opus where judgment matters, Sonnet where speed and cost matter, and human approval where accountability matters.

ChatGPT for All

Sep 30

∙

Preview

READ IN APP

I kept returning to the same uncomfortable question: if Claude Opus 5.5 is the stronger model, should it handle the entire project?

At first, that seemed sensible. Opus could understand the goal, create the plan, perform the research, write the deliverable, and review the result. One model held the context, so there were fewer handoffs and fewer chances for information to disappear.

But there was a hidden cost. Much of the project did not require difficult judgment. Once the plan was clear, many steps were structured, repeatable, and easy to verify. I was giving the most capable model routine work simply because it had started the job.

So I separated the roles. Opus would decide what needed to happen. Sonnet would complete the well-defined work. Opus would return only to inspect the finished result. A person would keep control of the decisions that carried real consequences.

That sounds simple, but the model switch was not the hardest part. The difficult part was handing work from one model to another without losing context, repeating completed steps, or letting a weak assumption travel through the entire project.

The useful insight is this: a two-model workflow succeeds because of the handoff system, not because two models are automatically better than one. Every stage needs a specific input, an expected artifact, a test, and a stopping rule.

This guide gives you the complete Planner, Worker, Reviewer, Approver workflow, including the project-state document, copy-ready prompts, quality gates, an illustrative cost comparison, and a reader-run test for deciding whether the system actually improves your work...

User's avatar

Continue reading this post for free in the Substack app

Claim my free post

Or upgrade your subscription. Upgrade to paid

Like

Comment

Restack

548 Market Street PMB 72296, San Francisco, CA 94104

Unsubscribe

Never let a signup go cold again.

Turn emails into revenue with Brew. No credit card, free credits to try.

I Made Claude Opus 5.5 Plan and Sonnet 5.5 to Work

substack.com · newsletter

Explore this email design and adapt it to your own brand. Review the copy, links, and offer before sending.

Email content

Forwarded this email? Subscribe here for more

I Made Claude Opus 5.5 Plan and Sonnet 5.5 to Work

A practical system for using Opus where judgment matters, Sonnet where speed and cost matter, and human approval where accountability matters.

ChatGPT for All

Sep 30

∙

Preview

READ IN APP

I kept returning to the same uncomfortable question: if Claude Opus 5.5 is the stronger model, should it handle the entire project?

At first, that seemed sensible. Opus could understand the goal, create the plan, perform the research, write the deliverable, and review the result. One model held the context, so there were fewer handoffs and fewer chances for information to disappear.

But there was a hidden cost. Much of the project did not require difficult judgment. Once the plan was clear, many steps were structured, repeatable, and easy to verify. I was giving the most capable model routine work simply because it had started the job.

So I separated the roles. Opus would decide what needed to happen. Sonnet would complete the well-defined work. Opus would return only to inspect the finished result. A person would keep control of the decisions that carried real consequences.

That sounds simple, but the model switch was not the hardest part. The difficult part was handing work from one model to another without losing context, repeating completed steps, or letting a weak assumption travel through the entire project.

The useful insight is this: a two-model workflow succeeds because of the handoff system, not because two models are automatically better than one. Every stage needs a specific input, an expected artifact, a test, and a stopping rule.

This guide gives you the complete Planner, Worker, Reviewer, Approver workflow, including the project-state document, copy-ready prompts, quality gates, an illustrative cost comparison, and a reader-run test for deciding whether the system actually improves your work...

User's avatar

Continue reading this post for free in the Substack app

Claim my free post

Or upgrade your subscription. Upgrade to paid

Like

Comment

Restack

548 Market Street PMB 72296, San Francisco, CA 94104

Unsubscribe

Get the appStart writing

Browse more email designs