bem_hq boosts agent accuracy 3%→100%

- bem_hq used 300+ agent simulations to refine API design and found that a single JSON example in docs lifted agent accuracy from 3% to 100%. - They also measured that returning complete data on the first API call reduced token consumption by about 35%, cutting costs for LLM-driven workflows. - The results argue for simulation-driven API design when embedding agents to improve reliability and lower token spend. (x.com)