Simultaneous claims that 'agentic' models are dramatically less efficient, but also forecasts efficiency improvements? We're in full-on tea-leaves-reading mode.