API Blog Sign in

AI agent

The AI agent is an AI teammate that works tickets alongside your team. It reads the conversation, finds the relevant answer in your knowledge, and drafts a reply — and, within the limits you set, it can send that reply and take actions on your behalf, or hand the ticket to a human when it shouldn’t act alone.

How it works on a ticket

  1. Reads the conversation and understands what the customer is asking.
  2. Grounds its answer in your brand’s knowledge — it answers from what you’ve given it, not from guesswork.
  3. Drafts a reply, and any actions it wants to take (like looking up an order or issuing a refund).
  4. Sends or parks — if the reply and actions are within your guardrails, it proceeds; otherwise it parks for your approval.
  5. Escalates — when it hits something outside its knowledge or authority, it stands down and hands the ticket to a person, with the context of what it found.

Shaping how it sounds

Under the agent’s Behavior settings you set its persona, a tone descriptor (with suggested-tone chips), and brand-specific instructions that are added to its base prompt — so it sounds like your brand, not a generic bot.

Test before you trust it

The agent is configured per brand under Settings → AI → Agent, where you can:

  • Reply test — see the reply it would draft for a given message.
  • Agent sandbox — run a full end-to-end conversation against it.
  • Observability — watch its real runs to see what it did and why.

Start it in a supervised mode (drafts only, everything parks), watch it in Observability, and widen its authority as you build trust.

Next steps

© Copyright 2026. All rights reserved.