This Huge Update Changed The Way I Use Claude Code

Anthropic's 'adviser strategy' for Claude Code uses Sonnet as the main executive agent for all code changes and tool calls, while Opus acts as a consultant only when Sonnet gets stuck. This approach outperforms Sonnet alone on benchmarks and costs significantly less than running Opus as the primary agent. Practical tests show it works well for simple-to-medium complexity apps, resolving stubborn bugs faster and saving tokens. However, limitations exist: Sonnet sometimes misjudges task complexity and skips consulting the adviser, and for large-scale or highly complex apps with many dependencies, using Opus directly as the main agent is still faster and more reliable.