Model "shallowness" is a big deal in the time of AI agents, models can be very good in narrow areas but since they are shallow, they lack context and reasoning to make good judgement calls when doing tasks. Once you are operating independently, being good at coding isn't enough