build1 publisher
Uber's AI cost per session fell 18 points further than its cost per request
The gap implies each session now uses roughly a quarter fewer requests, so the compaction threshold and the prompt cache in Uber's own harness are carrying part of a saving usually credited to cheaper open models.
Publishers:blog.pragmaticengineer.com
Reality
- Evidence45
- Adoption62
- Hype gap+26
- Incentives58
- Confidence48