Maybe hardware? Agents workflow design? Faster/more accurate models? Using our preferences to decide what to build and our agency to actually do it (poetically speaking, being the gradient in the agent optimisation process) ?

I don't know! But maybe that's how it feels like to be far from excellence again; maybe that's how it feels like to be out of our depth again, just like we were at the beginning of our career?

I don't want to risk stretching the analogy too far, and it might be that agents make all cognitive work obsolete. But surely there'll be something that defines a capability frontier? The alternative position is to believe that there just won't be a frontier anymore (I guess that's the RSI/AGI/ASI position :shrug)