Either the crap truly works and nobody is needed or it does not work. Where is this half-arsed human-agent hybrid vision coming from? The land of plateaued LLM gains?