|
|
|
|
|
by adam_rida
5 days ago
|
|
This feels like the natural next phase. A lot of companies went from “use as much AI as possible” to suddenly realizing that not every call needs frontier-model intelligence. In production, many calls are just repeated classification, tagging, routing, extraction, etc. The interesting problem is not banning usage. It is figuring out which parts of the traffic are safe to handle cheaply, and which parts should still go to the expensive model. |
|