"The next scaling law is the agentic scaling law. It's like multiplying AI — we could spin off agents as fast as you want. I have four scaling laws: pre-training, post-training, test time, and agentic."