
Set up an email agent with current guidance
Zendesk's current creation flow starts in the AI agents workspace: select Dashboard, Create AI agent, then for Email. Email includes API and web form channels. Select a brand and knowledge base; a crawler is optional, but Zendesk requires at least one connected knowledge source. Set a factual business profile, tone, default language and translations. On Set up your AI agent on email and web form, write the escalation acknowledgement, then choose Save and close or Go to activation, select addresses or forms, and click Activate on channels. Only one AI agent can be active for a support email address or web form.
This is not the old Essential or pre-March-2026 Advanced setup. The new product treats Email as its own channel. A messaging flow, especially availability behavior, should not be copied into an email workflow without checking what customers and staff will actually receive.
Prepare knowledge and escalation rules
Email is often where customers include account detail, a complex history, or a request that cannot be safely completed from a public policy. Start with a small source set: current policy pages, instructions the support team already uses, and only content that is safe to cite to a customer. Set an explicit boundary for identity checks, account changes, refunds, legal questions, and human requests.
For each boundary, name the next step. A failed lookup should not produce an invented answer. An escalation reply should not promise “soon” unless the team can meet that promise. Verify the receiving queue and ticket state; a fallback message does not itself mean a staffed person owns the case.
Test a real email job
Test the actual email channel with an expected result written in advance. Include a clear policy question, an incomplete request, a request for a person, a case that requires identity verification, and a deliberately failed action. Inspect the outgoing reply, sender identity, ticket fields, action log, escalation route, and the time it would take your team to respond. Re-run those cases after changing a source, instruction, action permission, or routing rule.
Review invoice replies with eesel CLI
eesel's CLI lets a support lead, a script, or a coding agent such as Claude Code operate the same teammate and workspace as the dashboard. For email support, use it to check whether the teammate asks for the right identifier before discussing an invoice.
Start with an existing teammate selected explicitly and an owner-approved fictional case. Inspect its instructions, connected sources, and action permissions before chatting: a CLI conversation is real, billed work. The following prompt requests a draft only, but that instruction does not replace permission controls.
npx @eesel/cli instructions --agent zendesk-email
npx @eesel/cli new --name "missing-invoice-email" --agent zendesk-email
npx @eesel/cli chat "Fictional review case: a customer asks for a March invoice but has not supplied an account reference. Draft a reply that asks for the minimum identifier required by our approved policy. Flag any missing policy instead of inventing one. Do not retrieve an invoice, send email, or change a ticket." --agent zendesk-email
Node.js 18.17+ is required. Commands return JSON, so Claude Code can compare the proposed reply with the approved identity policy and prepare a short change request: the sentence that failed, the source it should follow, and the expected replacement. A human checks that it neither discloses billing information nor promises delivery before verification.
If the owner approves an instruction change, inspect its command help and preview the write with --dry-run where supported. Confirm the applied change, rerun the fictional case, and include a second case where identity has already been verified. Use activity to inspect what the teammate actually did; approvals lists only actions currently held for a person.
Finish with a controlled test through Zendesk Email. Check the sender, delivered message, ticket fields, escalation acknowledgement, and receiving team. A good terminal draft alone does not prove that email delivery or routing works.
Try an AI teammate for Zendesk email
eesel can be the AI helpdesk teammate in the Zendesk workflow your team already uses. Its CLI is another way into the same teammate as the dashboard, which helps teams document a test before widening the job. Try eesel

Frequently Asked Questions
What are Zendesk AI agents in email channels?
They are channel-specific AI agents that can respond to customer email using connected knowledge and configured rules.
Is email set up the same as Messaging?
No. Zendesk's new AI agents are single-channel, and some conversation controls such as Messaging availability do not apply to Email.
What knowledge should an email AI agent use?
Start with a small, approved set of current sources and remove material that is expired, contradictory, or unsuitable for customer replies.
How should teams test email AI agents?
Use representative email cases and inspect the reply, ticket state, action result, escalation path, and operational response promise.




