How to choose Opus or Sonnet for Claude Code sub-agents

The author proposes three questions to guide the decision. First, does the agent determine strategy or judge quality? If yes, use Opus, as errors here derail subsequent work. Second, will customers read the text directly? If yes, use Opus to avoid awkwardness. Third, is the process fixed and checked later? If yes, Sonnet is sufficient. The author notes that even for writing tasks, Sonnet is acceptable if a manager reviews the output before use.

The distribution of the 16 employees shows 9 assigned to Opus, including a COO, corporate planning, managers, and a writing specialist. Seven are assigned to Sonnet, including a secretary, sales, and operations roles. The author explains that using Opus for everyone is inefficient due to speed and cost, while Sonnet offers better headroom for routine tasks. The key takeaway is that organizational checking mechanisms matter more than raw model intelligence for lower-stakes tasks.