Use Sonnet for Grunt Work, Opus for Architecture
Use Sonnet for Grunt Work, Opus for Architecture
Running Opus for every task is like taking a helicopter to the grocery store. It works, but you're burning resources on something that doesn't need them.
The pattern that actually optimizes your AI subscription: use cheaper, faster models for routine tasks and save the expensive models for work that requires deep reasoning.
What Sonnet Handles Fine
Renaming variables across a codebase. Writing boilerplate tests. Generating TypeScript interfaces from a JSON schema. Formatting data. Writing commit messages. Updating documentation.
These tasks are well-defined, have clear inputs and outputs, and don't require the model to hold complex architectural context. Sonnet handles them as well as Opus, often faster, and at a fraction of the token cost.
Where Opus Earns Its Keep
Designing the authentication flow for a multi-tenant system. Refactoring a tightly coupled module without breaking downstream consumers. Debugging a race condition that only appears under load. Planning a database migration strategy.
These tasks require the model to reason about tradeoffs, hold multiple constraints in memory, and produce solutions that are correct across edge cases. This is where the capability gap between model tiers becomes obvious.
The Practical Workflow
Start your day by triaging tasks into tiers. Batch the routine work and run it through Sonnet. Queue the architectural decisions for Opus sessions where you'll engage deeply with the output.
This isn't about being cheap. It's about not exhausting your Opus quota on tasks that don't benefit from it. When you actually need Opus at 4 PM for a critical design decision, you want tokens available - not burned on morning boilerplate.
The developers getting the most value from their AI subscriptions aren't the ones on the highest tier. They're the ones routing tasks to the right model.
Fazm is an open source macOS AI agent. Open source on GitHub.