The smarter the models, the less you have to structure the workflow. If you have a workflow that works well with a model today, you are in a good position because in a few months time you can switch to a much cheaper model.

One strategy I never had luck with was to have the smart model do an eval loop where it instructs the smaller model until it does the task. The problem is that the large model is way to prescriptive. This will improve in the near future because model labs are very focused on subagents