Skip to slide
Chapter 14 · Cost, Latency, and Model Tiering
141 / 191

CHAPTER 14 · Cost, Latency, and Model Tiering · 9 / 13

Stream to mask latency

Streaming (Chapter 8) doesn't reduce real latency but transforms perceived latency. A turn that takes fifteen seconds feels responsive when the user sees reasoning, then tool activity, then the answer flowing in. Showing meaningful intermediate events ("reading contract.pdf") turns dead time into visible progress. This is a cost-free latency win and one of the best UX investments you can make.

← → arrow keys work too