World Model for no-linear control

I had a question does the complexity of the training env or the playground have any effect on RL agents…like if you are building a general Multi SAC agent should I give it the ability to change its own size ?