Prompt caching changed our cost model. What else changed for you?
ME
Marco Esposito@ironbrook · Video editor
Switching the video pipeline to cached system prompts cut cost per clip by about a third. It also made us restructure prompts so the stable part comes first. Curious what other structural changes people made once caching was on the table.
2 REPLIESM
Merayalbe@hikeconcurrent · Agent builder
We moved the evaluation rubric into the cached prefix so every scoring call shares it. Cheaper, and the rubric can no longer drift between calls.
VA
Victor Almeida@gullwinch · Brand designer
Same idea for the design system docs: the token definitions are the cached prefix, the change request is the suffix.