Channelsintermediate

Route an email channel to an Agent Stack

Set an email channel's Route To to AI Agents so the orchestrator answers inbound mail. Covers the two reply-delay floors and the guard that stops the AI from replying to autoresponders, OOO messages, bounces, and spam.

6 min read

Route an email channel to an Agent Stack

Email channels can hand inbound mail to an Agent Stack the same way Web Chat and Messenger do. Once a channel’s Route To is set to AI Agents, every inbound email runs through the orchestrator, which drafts and queues the reply instead of waiting in the team inbox for a human.

This article covers the email-specific behavior you should understand before flipping the switch: the reply-delay floors and the guard against auto-replying to automated mail.

Before you start

  • The email domain must already be added and verified.
  • You need an Agent Stack configured and ready to take conversations — content routing and handover logic live in the stack itself and are unchanged by anything in this article.
  • You’ll need admin access to Settings → Email.

Set Route To on the channel

  1. Go to Settings → Email and select the channel you want to route.
  2. Open the Route To picker. Alongside your teams, you’ll now see AI Agents as an option.
  3. Select the Agent Stack you want handling this channel’s inbound mail.
  4. Save. From this point on, inbound email to this channel’s address is handed to the orchestrator instead of (or before) landing with your team.

The panel description updates to reflect this: incoming emails are routed to the selected team or AI agent stack, whichever you picked.

Reply-delay floors

When a channel is routed to an Agent Stack, a new AI reply delay section appears with two fields:

  • First reply — floor for the very first AI reply on a conversation. Defaults to 180 seconds.
  • Follow-up replies — floor for every reply after that. Defaults to 90 seconds.

Both are measured from receipt of the inbound email, not from when the AI finishes drafting. That distinction matters: the floor absorbs however long the AI takes to work, rather than stacking on top of it. If the AI finishes in 10 seconds, the reply still waits out the rest of the floor. If the AI takes 200 seconds on the first message, the reply goes out as soon as it’s ready — the floor has already been absorbed by the work itself.

The point of the floor is to keep replies from reading as instant auto-acknowledgments. A response that lands in the recipient’s inbox one second after they hit send feels like a bot, even when the content is good. A short, deliberate pause reads as a real reply.

Under the hood, this delay is delivered via the outbound queue’s native delay mechanism — no worker sits blocked waiting for the clock to run out.

Each value accepts 0–3600 seconds. Setting a field to 0 removes that floor entirely; the reply goes out as soon as the AI is done. You will typically want the follow-up floor lower than the initial floor, since a fast back-and-forth on an established thread reads more naturally than a fast first response.

The automated-mail guard

The orchestrator will not auto-reply to:

  • Autoresponder acknowledgments
  • Out-of-office (OOO) messages
  • Bounces
  • Spam

This guard reuses the RFC 3834 auto-submitted signal to detect automated mail before the AI ever drafts a reply. It’s not something you configure per channel — it applies automatically to any channel routed to an Agent Stack, so an OOO reply to your support address doesn’t spawn an endless exchange of auto-replies.

Stand-down and give-up still work the same way

If a human joins the conversation (stand-down) or the Agent Stack hands the conversation off to a human (give-up), email follows the same channel-agnostic behavior as every other AI-routed channel. There’s no email-specific override to configure — whatever your Agent Stack’s handover rules are, they apply here unchanged.

Verify it’s working

  1. Send a test email to the channel’s address from an external mailbox.
  2. Confirm the conversation shows the Agent Stack as the responder in Conversations.
  3. Time the reply — it should land no sooner than your First reply floor, and the delay should visibly absorb whatever time the AI needed to draft.
  4. Send a reply to that thread and confirm the follow-up delay (not the initial one) applies.
  5. Optionally, send a message with an out-of-office auto-reply header (or reply from an account with OOO active) and confirm no AI reply is generated.

See also

Tags

How ToConfigurationAgent Stacks