At the moment, AI FDE doesn’t let you switch models when launching subagents.
Concrete example: when building an OSDK React app, I’d like to do the research and design with Opus and produce a well-structured Epic with sprint notebooks — but I can’t then launch a subagent to implement Sprint 1 with Sonnet.
Being able to choose the model per subagent would help a lot. It would keep token consumption (and cost) down for the implementation-heavy work, and it would remove the manual session switching that currently breaks the flow.