In many cases the AI FDE sessions take the model that I’m using as a model for the subagents, which is kind of a problem sometimes. If we could really get an option to configure our subagents even better, like choosing a different model prior to the start of sessions and having some kind of configurability exclusive to subagents, that will help us to write the code. It will be much less verbose and can handle a lot of problems effectively using a different model that we wish we could use for a subagent, without going and changing it while the session gets started. When starting the work of the session
We recently introduced the concept of model categories to AI FDE (smart, balanced, and fast). The main model chooses a category of model to use based on the subagent task. So for example, if the subagent being spawned is doing very simple work, the main model should choose a fast category model for that subagent. If you find that the main model is choosing a category you believe does not match the subagent task, one thing you could do is create an AIP skill which provides guidance on which model categories to use for which bodies of subagent work.