TLDR: to complete real complex world tasks like autonomous product development, you need the context from each previous subtask. So you can't really run subagents in parallel. Yet this stresses the context, so you need a way of extracting the key context from each subtask for use with all subsequent subtasks.
TLDR: to complete real complex world tasks like autonomous product development, you need the context from each previous subtask. So you can't really run subagents in parallel. Yet this stresses the context, so you need a way of extracting the key context from each subtask for use with all subsequent subtasks.