I agree. We need a harness that can choose which model to use: sometimes an open model, sometimes ChatGPT. With Claude, for example, the system may be incentivized to use the most expensive model. We need an open-source counterpart so we are not locked into one model.