I believe that at some point in the next ten years, a "brain computer interface" will exist that detects some kind of neuron activity, runs it through a machine learning model, and translates it into LLM agent actions. It won't be anything as clean as "you think a sentence and it types it". It will be something one level lower than "prompting an AI". You do some vague mental thing and it translates into a result on the computer. I believe this product will not work, but be sold anyway. (1/2)