Not to sound like an LLM, but that seems exactly right to me. Use it as a cheaper, high-functioning task subagent and lower reasoning for a master Opus session. As long as not every portion of your task requires maximum intelligence, you should come out ahead.