Loading page...Загружаем страницу...正在加载页面...
AtlasRepo

My AI cron job looked cheap at 5 runs/day. At 96 runs/day it turned into 1,728 model calls

Описание и применение

В тестах AI‑cron‑задача казалась дешевой, но в продакшене из‑за повторов, опросов и нескольких окружений количество вызовов модели выросло до 1 728 в день, что значительно увеличило расходы.

Задача

AI cron jobs that seem inexpensive in testing can explode in cost in production due to retries, polling, multi‑step chains, and multiple environments.

Результат

Estimating cost by counting model calls per run, retries, and environments provides a realistic budget. Switching to flat‑rate compute removes surprise usage bills.

Рабочий процесс

  • Schedule cron every 15 minutes
  • Fetch new records
  • Call GPT‑5.4 for classification
  • Call Claude Opus 4.6 for summarization
  • Retry on timeout
  • Poll downstream status
  • Send output
  • Repeat in dev, staging, and prod environments

Доказательства

96 runs/day × 3 model calls/run × 2 retry multiplier × 3 environments = 1,728 calls/day

Источник

The expensive part of an AI cron job usually isn't the prompt.

Источник

  • Дата записи: 2026-09-22T10:45:37.631Z