Run customer support with AI agents and human approval
For: small teams who want fast first responses without letting a bot talk to customers unchecked. You get: every request as a ticket, an agent that does the first pass, and a person who signs off on each customer reply.
The setup
| Step | Where | Why |
|---|---|---|
| 1. Open Support | Modules → Support | A branded portal. Each request becomes a case and a ticket. |
| 2. Choose the intake lane | Modules → Support: pick a board and lane | All new requests start in one lane. |
| 3. Connect an agent | Workspace Settings → Agent Connections | Give it projects:read and projects:write and a grant for this project only. |
| 4. Keep replies human-approved | Agent replies to customers need approval (on by default) | An agent's customer reply waits in AIr Traffic Control until a project admin approves it. |
| 5. Add rules | Workspace Settings → Automations | For example, set a priority when a ticket enters the intake lane, or archive tickets 30 days after they are complete. |
How it runs day to day
- A customer writes in through the portal.
- The agent reads the new case, sets the priority, adds a private note with what it found, and prepares a reply.
- A project admin reviews the reply in AIr Traffic Control and approves it. Nothing reaches the customer before that.
- If the case needs product work, the agent or a person creates a child Bug or Task from the Support tab. It stays linked to the case.
- When the fix ships, reply to the customer and resolve the case. ToDoddle blocks completing the ticket while the conversation is still open.
Tips
- Private notes are never shown to customers. They are where the agent and the team think out loud.
- Use Mark as spam or Block sender for abuse. Both can be undone.
- Agents only see projects you grant. Start with one project.
Related
Updated 2026-11-10. Owned by product.

