My AI cron job looked cheap at 5 runs/day. At 96 runs/day it turned into 1,728 model calls
Description and use
In testing, an AI cron job appeared inexpensive, but in production, retries, polling, multi‑step chains, and multiple environments inflated model calls to 1,728 per day, driving up costs.
Problem
AI cron jobs that seem inexpensive in testing can explode in cost in production due to retries, polling, multi‑step chains, and multiple environments.
Outcome
Estimating cost by counting model calls per run, retries, and environments provides a realistic budget. Switching to flat‑rate compute removes surprise usage bills.
Workflow
- 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
Evidence
96 runs/day × 3 model calls/run × 2 retry multiplier × 3 environments = 1,728 calls/day
The expensive part of an AI cron job usually isn't the prompt.
- Record date: 2026-09-22T10:45:37.631Z