I’m currently using n8n version 1.107.4, and I’ve noticed a significant change in the behavior of manual node execution.
Previously, when testing a workflow step-by-step, I could manually execute individual nodes and n8n would simply use the data already produced by previous nodes (without re-running them).
However, in the current version, whenever I try to execute a node manually, n8n automatically re-executes the entire workflow from the beginning up to the selected node. This results in:
Unwanted API calls to earlier nodes (such as AI agents or third-party services).
Increased cost due to repeated calls to paid services (like LLMs, external APIs).
Slower debugging and reduced efficiency in the development workflow.
This new behavior might improve safety or consistency, but in practice, it’s counterproductive when working with long workflows and cost-sensitive services.