2 ms·

On our tests subagents & well used workflows are 20-30% more expensive for context & token efficiency