With an AI agent on your tickets, every new customer email is read by AI before anyone on your team opens it. The agent finds the customer's context, works out what is needed, writes an internal note that says who acts next, and prepares a reply draft. A person always checks and sends the reply.
This guide sets it up step by step. The agent itself is described in AI Agents: let AI work on your tickets.
1. Create the support agent
Open AI Agents (Tools menu) and create an agent, for example "Support". Choose provider and model (or leave empty for your default AI settings) and set it to Active. Keep Autonomy at Drafts only.
2. Write its instructions
The instructions are where the agent learns your business. Write them as you would brief a new colleague:
-
What you sell and who your customers are.
-
How to reply: language, tone, length, signature. Tip: "Only state what you have checked."
-
How to sort requests and what to do with each kind. Use your own categories, for example:
- Software company: Bug (something that should work does not: find the cause), Improvement (a new wish: note it for a decision, promise nothing), Question (answer it).
- Kitchen company: Complaint (book a service visit), New order (prepare a quote), Question.
- Association: Membership, Event, Board matter.
Ask the agent to write the category on the second line of its note, e.g. "Type: Complaint".
3. Describe your team
The agent's note starts with Next: <who> - <what to do>, so it needs to know who does what:
- Groups: under Settings > Groups, give each group with active users a short description of what it handles, e.g. "Installers: assembly and service visits".
- Other AI agents (optional): an active agent with a Description can receive work from the support agent. Write in the description what it does and how to hand work to it.
4. Choose which customers get AI
Use the field Incoming Message Automation on the organization (contacts get the organization's value when theirs is empty):
| Value | Meaning |
|---|---|
| Automatic | The agent runs on every new email from this customer. |
| Manual review | The agent also runs; use it for customers you want to watch more closely. |
| Block | Never automatically, e.g. spam or customers you do not want AI on. |
| Empty | Not automatically. |
Tip: set Automatic for customers you have already had a real dialog with; they are hardly spam.
5. Start the agent on new email
Create a workflow on Tickets, "Every time the record is saved":
- Conditions (all): Last Email At has changed, Waiting for answer from is Support, AI status is not Running.
- Conditions (any): the organization's Incoming Message Automation is Automatic or Manual review, or the contact's is.
- Action: Update fields: AI agent = your support agent and AI status = Start. Or use AI workflow action > Run AI agent.
You can always start the agent by hand with a comment that starts with @ai, also on customers set to Block.
6. Daily use
- Open the ticket and read the agent's note. The first line says who acts next.
- Click Insert in email on the reply draft, adjust it and send it.
- To steer the agent, write a comment starting with @ai, e.g. "@ai the customer is on the old plan, check that first".
Good to know
- The agent never sends anything to the customer itself.
- It only sees records its assigned user may see.
- Access to other systems (e.g. a customer's own system) needs a person's approval first.
- Each run uses your own AI provider's key and is billed by them.