Using subagents as parallel perspectives on design decision

Most of us use Claude Code subagents for parallel work - search this, refactor that, concurrent execution. Lately I've been using them for something different: parallel perspectives on the same problem. If you've discussed design decisions with the main agent long enough, you've hit this: the agent gets buried in the context, or the conversation grows long and it just starts agreeing with you. Either way, fresh angles stop coming. Subagents don't share that context. They come in cold - no history of the conversation, no accumulated framing - and that's exactly why they work here.