I don't want to beat a dead horse or regress, but in no way was Brett implying that hi-res NWP will/should replace forecasters. Not sure where you got that from. He was acknowledging how well CAMs seem to be performing this season, and that ignoring their consensus should be done at one's own peril (e.g., dewpoints in the 60s yesterday, characteristics of UH swaths from the past few threats, etc).
Of course epic fails will still occur given the extremely nonlinear nature of convective initiation and the occasional failure of parameterization schemes to capture things like surface fluxes during droughts, etc. Observations are still crucial and nothing can really be substituted for forecaster intuition and experience, but IMO we are beginning to enter a time where hi-res stuff offers insight that you can't get otherwise. Once a model is spun-up and the physics adjust and balance, it's totally reasonable to expect these models to have represent extremely small scale features that models with lower grid spacings and especially our current network of observations can't capture. Of course if all models are showing a certain UH swath that comes to verify it can be a bit of a black box problem to determine what exactly they all honed in on to cause that. But given the relatively low density of our observing networks, and assuming proper analysis and quality control of the data fed in, I think one has to at least strongly consider the immense value that a model data assimilation system (that is, the optimal combination of both the observation networks and the model's physical constraints, carrying information from past assimilation cycles forward, and having a complete suite of physics (i.e. a soil model for surface fluxes, etc)) can offer as a forecasting tool.