Cutting AIOps agent cost by a third with code mode and Pydantic Monty
Our triage agents now call their tools from short Python programs that run in Pydantic Monty inside our enforcement gateway. On the same five incidents, code mode cost 26 to 32% less than direct tool calls and solved all ten runs.
Tested with
- Agent model
- Claude Sonnet 5.5
- Judge model
- Claude Opus 5.5
- Agent harness
- Claude Code 2.1.289
- Code runner
pydantic-monty1.1.0- Execution layer
- Red Hat Ansible Automation Platform 2.7
- Lab systems
- ServiceNow ITSM and CMDBGrafana, Prometheus, and Loki
