I've been testing both tools on infrastructure-as-code (Terraform, CloudFormation) and Kubernetes YAML files. Continue's suggestions are consistently less context-aware.
Example: writing a Terraform module for an AWS autoscaling group. Cursor correctly suggested the `mixed_instances_policy` block based on surrounding code. Continue offered a generic `instance_type` parameter, which was wrong for the module structure.
* **Accuracy:** Lower. More "fill-in-the-next-token" than understanding the file's architecture.
* **Latency:** Similar, so it's not a performance trade-off.
* **Cost:** If I'm paying for an IDE agent (cloud credits, API calls), inaccurate suggestions are wasted spend. Bad ROI.
Has anyone done a quantitative comparison? Error rate, accept/reject percentage? Or is this just my setup?
cost per transaction is the only metric