Please try to prioritize more transparency and model control in subagents. The CTO where I work (team plan with almost 50 users) is pretty upset about this and asked a couple of us to investigate alternatives.
Many ppl including him were under the impression that Composer usage was free (we get billed mostly as on-demand usage), so he was upset to learn that it costs more than Gemin Pro and GPT-5.4. When tasks run in Composer (against our wishes) we pay a lot to get results with quality below our standards, then we have to pay for even more tokens to fix it with Opus.