Work

AI support agent in Slack

Classifies inbound messages, answers from a knowledge base during working hours, and hands off when it should not speak.

  • I designed and built the routing logic, knowledge base, and escalation.
  • n8n
  • OpenAI/Claude
  • Slack
  • webhooks

Inbound Slack messages aren’t all the same job. Some are hours and policy. Some are account-specific. Some should never be answered by a model.

The agent classifies the message first. During working hours it answers from a business knowledge base when the class is safe. When it isn’t — account data, edge cases, anything the base doesn’t cover — it escalates to a human with the original thread intact.

I built it in n8n with OpenAI. The model call is the easy part. The work is the routing: what counts as an answerable class, what hours it’s allowed to speak, and how a handoff looks so the human doesn’t start from zero.

If the classifier is wrong, the system should fail closed — escalate — rather than invent a policy.

Start a conversationWork