LLM cost and latency engineering

SWE-Bench does not measure the job: why agent scores are the wrong readiness signal

A practitioner's field notes put curated agent evaluations on one axis and production intent resolution on another. The cost of confusing them gets billed per session, not per query.