Running agents with SLOs
The harness and its model deployed on Kubernetes with SLOs for time-to-first-token and task success, an alert, a load test, and a postmortem.
Builder · 2026–
What it is
The agent harness and its model deployed on Kubernetes and run like a service, with explicit SLOs for time to first token and task success.
How it's built
- Autoscaling for the model-serving workload.
- SLO dashboard tracking time to first token and task success, with an alert on each.
- Load test against the deployment, and a postmortem written up from it.
- Cost per completed task as the number that ties it together.
- Terraform is used only for what is actually deployed.