All the components necessary for AGI have already been developed.
The people who could conceivably “smush” all the necessary components for AGI together are all working on publishing incremental improvements. The incentives of academia are arguably setting back progress in a sort of convoluted, nerd-sniping way.
I think that we need cells of experimental engineering that focus on applying published research. OpenAI sort of does this, but they seem to be in a precarious position of having to answer to huge established corporate interests. This is like driving a tanker up a mountain.