The Insight
Most users pick a default Claude model and stay with it. A more productive approach is to treat Opus and Sonnet as two different specialists and route tasks accordingly.
Elizabeth Stief articulated the clearest version of this split during the February 5th session in response to Donald asking when Opus is substantially better than Sonnet:
“For me Opus is better in analysis, deep thought tasks, synthesis of data in different files, better research. Sonnet is better in writing/generating content. I use Sonnet for implementation of what I have constructed with Opus.”
The implication: Opus builds the architecture, Sonnet executes on it.
The Framework
Use Opus for:
- Analysis and synthesis of complex or multi-file data
- Deep research and literature review
- Strategic thinking, framework development, first-principles reasoning
- Tasks where you need genuine depth over fluency
Use Sonnet for:
- Content generation and writing
- Implementation of decisions already made
- Rapid iteration and output production
- Tasks where quality at speed matters more than maximum depth
The Practical Pattern
- Do your thinking with Opus — let it construct the architecture, the framework, or the analysis
- Export or summarize the key outputs
- Feed those outputs to Sonnet as a brief and generate the final content or implementation
This two-stage workflow extracts the depth of Opus without paying Opus-level token costs for the entire process.
Why This Matters for Coaches
Coaches and knowledge workers frequently conflate two different types of AI work: the generative-thinking phase (where a client’s situation is being mapped and synthesized) and the production phase (where content, proposals, or frameworks are being written up). These are different cognitive operations — and they benefit from different models.
The larger principle: knowing your tools at the level of their behavioral tendencies is as important as prompt technique. Model selection is a skill.