On Day 1, I learned that LLMs generate one token at a time. 22 days later, I realized users can actually see that loop happening. That's streaming. Same response. Same total time. Completely…