This is so interesting.

I'm unlocking savings (in both speed and costs) but using a different way of thinking by integrating Decision models:

In short, LLMs spend so many turns on discovery, triage, and safety checks for everyday tasks. A decision model cuts those turns by preparing the relevant context before the prompt and handling the simple judgments itself, so the LLM only does the real work.

Here is the full article I published yesterday with in-house benchmark: https://tawfeks.github.io/decisionkit/

What do you think, guys? I need feedback on this.