Question 1

What makes a good automation alert?

Accepted Answer

This looks simple but tests operational experience. "My criterion: the person reading the alert should know what to do. That needs four elements: 1. What happened — which workflow, which node, the error text 2. Scale — did one record fail, or all 300? That sets the priority 3. Context — which customer, which order ID 4. A link — to the execution, so they can see the detail in one click Just as important is when not to send. Alert on every small event and within two weeks everyone mutes the channel — and the genuinely critical message is lost with the rest. So my rule is: anything that demands no action is a log entry, not an alert. In practice I split them: critical failures go straight to the channel, recurring small problems go into a daily digest."

Question 2

How do you add a human approval step to an automation, and when is it needed?

Accepted Answer

Alongside the technical answer, explain when it is needed — that is where judgement is tested. "The technical side is simple: the messaging nodes' send-and-wait operation pauses the workflow, and once the answer arrives you branch with an IF. Telegram uses buttons, Slack an interactive message. The more interesting part is when it is needed. I have three criteria: 1. Irreversibility — payments, deletions, a message sent to a customer. A mistake cannot be taken back. 2. Cost of a mistake — when the amount or impact crosses a threshold. So I usually make approval conditional: automatic below 5000, approved above. The process stays fast while the risky cases stay supervised. 3. When AI made the decision — I put an approval in front of any irreversible action based on a model's output. One important technical nuance: a workflow waiting for approval is paused for a long time. So you also have to design what happens if no answer comes — a time limit and an escalation."

Question 3

How would you choose between Telegram and Slack?

Accepted Answer

This is a context question rather than a technical one — show that in your answer. "The choice depends mostly on where the team already is, not on node capabilities. An alert has to land where people already look; a message sent to a platform nobody opens is a message not sent. The technical differences are secondary but real: Telegram — a bot is minutes of work: a token from BotFather, a credential in n8n, done. It also suits talking to external users, a customer-facing bot for example. Slack — the corporate standard: channel structure, threads, user groups, integration with existing workflow. In exchange you have to create an app and grant scopes correctly, and some operations require a particular token type. In practice I often use both: internal team alerts to Slack, external user interaction (bots, intake) to Telegram. I extract those into a sub-workflow so the decision about which channel to use lives in one place."

Scenario	How it is built
Critical failure alert	Error Trigger → Edit Fields (compose the message) → Slack/Telegram
Daily digest	Schedule → data sources → Aggregate → one message
An operation needing approval	Trigger → a Slack/Telegram "send and wait" step → IF on the answer → the operation
A chat bot	Telegram Trigger → AI Agent (with Memory) → Telegram (Send Message)
Creating a task from a channel	Slack Trigger (a reaction) → Jira/Notion → Slack (reply in thread)

Telegram, Slack and notifications

📚 Sources and documentation