
What Essential did, and how to identify your setup
Essential provided generative answers from connected knowledge. That remains a useful job: answer documented questions and give customers an appropriate next step when the material is insufficient. But neither a published article nor a generated answer guarantees accuracy. An outdated returns rule is still outdated when an AI quotes it.
To identify your configuration, open Admin Center > AI > AI agents > AI agents, then inspect an agent under Essential (legacy). Essential agents lack Intents and Answers tabs; legacy bot-builder agents have Answers and may have Intents. The migration guide explains these distinctions.
Make a working inventory: agent name, channel, source owner, languages, escalation route, and the person approving replacement. Record the customer jobs, not only the agent's settings. “Answer delivery questions” may hide different policies for domestic orders, international orders, and delayed shipments.
What changes in the replacement
| Existing setup | What to plan |
|---|---|
| One Essential agent on messaging and email | Separate new agents for each channel |
| Standard responses | Corresponding system replies, plus failure cases |
| Essential configuration in Admin Center | Ongoing configuration in the AI agents workspace |
| Ongoing conversations during activation | A handover where the replacement has no prior conversation memory |
These are documented migration differences, not a promise of an automatic copy. Zendesk's guide still describes a migration tool as coming later in 2026; confirm availability in your account. Manual recreation is documented. Plan activation during a quieter period and prepare agents for customers who repeat context.
Do not add complex actions simply because the new experience supports them. First reproduce the useful answer and handoff behavior. Treat an order change or refund action as a separate permission and testing decision.
Create the new knowledge-based setup
For the general current experience, a client admin opens Dashboard > Create AI agent in the AI agents workspace and selects Messaging or Email. Choose the brand and knowledge base; an external website crawler is another supported source. At least one knowledge source is required to continue.
Then review the name, business profile, tone, languages, and channel responses. Keep business-profile text factual rather than using it as an instruction field. Save and close saves configuration without activating it; Go to activation makes it available on the selected channels. Follow the current creation guide, which distinguishes this flow from the one for pre-March 10, 2026 Advanced customers.
Before activation, ask the policy owner to inspect several cases:
- A normal question with a clear answer in the approved source.
- An exception the source does not resolve.
- A customer asking directly for a person.
- A failed lookup or unavailable handoff.
- The same question in each supported language.
Check the answer and source together. Then inspect the actual customer channel, receiving queue, and resulting ticket. A plausible preview does not prove routing or account access is correct.
Recheck the budget instead of carrying forward old prices
The old shorthand “every deflection costs about $2” is not a reliable budget. Zendesk's resolution tiers distinguish assisted escalations, contained resolutions, and verified resolutions. Assisted and contained tiers do not draw against the resolution allowance; verified resolutions do. Verification happens after the conversation ends, with timing that varies by channel.
Ask the account owner to confirm the allowance, applicable price, and overage controls. Estimate the volume of the specific job being enabled, then compare actual usage after rollout. More automation can change usage even when the existing overage settings remain in place. Do not call all unanswered follow-ups successful resolutions.
Use eesel CLI to check what your support knowledge covers
A support lead can ask Claude Code, Codex, or Cursor to review an eesel teammate's readiness for the same customer jobs. The eesel CLI gives the coding agent, a person in the terminal, or a script access to the same teammate and workspace as the eesel dashboard.
For example, prepare an approved, non-sensitive checklist of delivery questions and expected policy references. Get permission to upload it and inspect the teammate's connected action permissions first. Then check source-download status and ask for a gap review:
npx @eesel/cli --agent support integrations download list
npx @eesel/cli --agent support files upload ./delivery-coverage-checklist.md --dry-run
npx @eesel/cli --agent support files upload ./delivery-coverage-checklist.md
npx @eesel/cli --agent support new --name "delivery-coverage-review"
npx @eesel/cli --agent support chat "Review the uploaded delivery checklist against your approved knowledge. For each question, identify supporting sources or missing information and propose a clarification or human route. Do not send customer messages or change orders."
With Node.js 18.17 or newer, these commands return JSON. A script can retain the response in an approved review record; the coding agent can compare the cited material with the checklist and propose source or instruction changes. The support owner should open the citations, correct any mistaken gap, approve the proposed changes, and repeat the affected questions.
This workflow checks the configured eesel teammate. It does not import or replace a native Zendesk Essential agent. Uploads change the real eesel workspace, and chat is real work rather than an isolated simulation. A request not to act is not a permission control; held approvals cover only actions actually placed in that queue.
eesel's current pricing uses task-based usage, not the old flat monthly interaction-bundle description. The public rate for a regular support ticket or chat session is $0.40, covering the interaction rather than each reply. Tasks are billed regardless of outcome, so include review and testing in your budget rather than assuming only successful answers cost money.
Keep the support job clear while the product changes
Replace the affected Zendesk configuration because the legacy product is retiring. Evaluate eesel when you also want a teammate whose knowledge and work you can inspect from your terminal or coding agent. These are related decisions, but they are not the same migration.

Try eesel with one approved support job. Use the CLI to inspect the sources and review an answer, then confirm the real customer experience before expanding access.
Frequently asked questions
What was Zendesk AI Agents Essential?
It was a previous level of Zendesk AI-agent functionality for answering customer questions from connected knowledge. It is being replaced by the new AI agents experience.
When is Zendesk AI Agents Essential being removed?
Zendesk lists December 10, 2026 as the removal date for Essential and legacy functionality. Maintenance mode began August 31, 2026.
Does the removal notice apply to existing Advanced customers?
The notice explicitly excludes customers who previously purchased the AI agents - Advanced add-on. Do not confuse a packaging change with a requirement to rebuild every Advanced agent.
Can one replacement agent handle messaging and email?
New agents are single-channel. An Essential deployment covering both needs separate messaging and email agents, with each channel tested before activation.
Does help-center content guarantee correct AI answers?
No. A source can be outdated, incomplete, or misapplied. Check representative answers, missing information, and human handoffs against approved policy.
How can eesel CLI help during a support review?
A person, script, or coding agent can inspect the same eesel teammate and workspace as the dashboard, check source status, and ask it to identify gaps against an approved support checklist. Review permissions first and verify the result; this does not migrate native Zendesk agents.






