Key Info

Ahmad Awais says CommandCodeAI's internal RSI engine treats "cache" as its north star metric, a choice he credits with enabling $100 worth of inference for $10 at a scale of 100+ trillion tokens.

Highlights

  • The team runs a decision model at 100+ trillion token scale and finds 2-3M code improvements daily.
  • Cache is framed as the highest-quality "fuel": the more you use, the more mileage and refined output you get.
  • Awais argues lines of code is a poor metric; the better metric is one the end user actually cares about.
  • The cache metric is described specifically as the north star for the internal RSI engine, not the whole company.