n8n Agents: let AI choose the next step, and workflows define its execution
n8n connects applications and business processes. Agents let teams pursue cross-tool goals while configured workflows, approvals and execution records keep concrete actions under control.
Form your own approach first, then explore the reasoning to check trade-offs and validation you may have missed. There is no single right answer.
What should a person approve before an Agent escalates a case?
Scenario
A support agent has prepared a customer summary for an on-call channel. Sending exposes the information to another team, so a person must confirm first. Design an approval interface that makes the customer, destination and message clear, and explains that approval covers this action rather than all future sends.
Given conditions and requirements
Design an approval screen for a support Agent's “escalate to the on-call team” action.
- The Agent has looked up a customer and is about to send a summary to an on-call channel. The person reviewing it may be handling several customers.
- Each approval applies only to the current customer, destination and summary. Changing any field after approval requires a new decision.
- The user can approve, reject or return to the conversation to provide more information. Do not show a modal for every read-only lookup.
- Explain how the interface shows context, impact, invalidated approval after edits, and feedback after repeated clicks. Propose tasks that reveal mistaken approval. Do not assume this is n8n's existing interface.
Work through this case
The panel initially shows customer A, channel X and summary version 1. While the user reads it, the summary changes to version 2 or the destination becomes channel Y. Show how the previous approval becomes invalid, how the change is explained and what must be visible before approval again.
What to deliver
Deliver sketches or a state table for review, changed content, execution and an unknown outcome, plus key approval/rejection feedback and a task that reveals mistaken approval. Agent retrieval and reasoning algorithms are out of scope.
Explore the reasoning
- Lead with the effect. Show the customer, destination channel, proposed message and reason for escalation. Make supporting evidence expandable; a long log should not be the default view.
- Bind approval to a specific version. Changing the customer or message invalidates approval and exposes the change. “Allow escalation” must not become permanent permission for any future escalation.
- Separate rejection, missing information and execution. Rejection cannot appear as a sent notification. Repeated clicks after execution should return to the same result. An unknown outcome stays pending confirmation rather than appearing undone.
- Observe decisions. Give participants two similar customers, a destination change and a timed-out response. Watch whether they notice the changed target and mistake uncertainty for failure. Draw conclusions only after running the test.