Opus 4.6 Is Production-Ready - But Only If You Write the Spec First
The debate about whether Opus 4.6 is ready for production-level code misses the point. The model is not the bottleneck - your specification is.
1,500 Font Calls Migrated in One Session
We had Opus 4.6 migrate over 1,500 font calls across an entire SwiftUI codebase for a global font scaling feature. Regex patterns, mass replacement, settings UI - all in one go. It worked because we wrote a detailed CLAUDE.md spec before the agent touched anything.
The spec included exact file paths, the specific regex patterns to match, the replacement format, and explicit acceptance criteria. The agent was not making design decisions - it was executing a well-defined plan.
The Spec Is the Engineering
Without a spec, Opus fumbles. It guesses at file locations, makes assumptions about naming conventions, and produces inconsistent results. This is not a model limitation - it is a context limitation.
When you write a detailed spec with:
- Exact file paths to modify
- Clear input/output examples
- Acceptance criteria for each change
- Edge cases to handle
The model stops guessing and starts executing. The spec is doing the engineering. Opus is doing the typing.
Why This Works for SwiftUI
SwiftUI codebases are particularly well-suited to this approach because the patterns are consistent. Font calls, color definitions, spacing values - these follow predictable structures that a well-written spec can capture completely.
The same approach works for any repetitive refactoring task. Database migration patterns, API endpoint updates, test fixture generation - anything where the transformation rules are clear and the scope is defined.
The Real Skill Shift
The skill that matters now is not writing code - it is writing specs that produce correct code. Engineers who can decompose a problem into precise, unambiguous instructions for an AI agent will outperform those who try to use the model as a pair programmer without preparation.
Fazm is an open source macOS AI agent. Open source on GitHub.