I see cost control as one of the most important parts of automation.
It is not enough to automate something. You also need to understand what it costs to run.
Testing n8n AI Assistant
Building a working workflow with AI
I tested n8n’s new AI Assistant by connecting it to OpenAI (GPT-5.6 Sol) and setting up n8n’s Sandbox environment.
I asked the Assistant to:
Generate 5 random numbers and calculate their average.
The Assistant created the workflow directly in n8n. The workflow generates five numbers using JavaScript and calculates their average using an Aggregate node.
Result:
5 numbers → Average: 50.4
AI creation cost:
$0.74
The $0.74 represents the AI/API cost of creating the workflow. The resulting workflow does not require an LLM to perform the calculation, so there is no recurring AI cost for the calculation itself.
My take on the cost
$0.74 is relatively high for such a simple workflow.
The workflow itself is trivial: five random numbers and one average calculation. The interesting part is therefore not the computational cost, but the cost of having an AI agent understand the request, interact with n8n and build the workflow.