Architecture + Claude Code + Performance

Why expensive models make terrible execution subagents

Subagent model selection has one costly default. Reasoning models obey their own traces 15.6% of the time, so here is where to pin model and effort.