Zendesk AI Agents Advanced: what changed in 2026

What changed, and what did not

Zendesk says it began rolling out the new packaging between May 11 and June 12, 2026. The new experience expands access to capabilities previously offered through AI Agents Advanced, including dialogue builder, use cases, generative procedures, and richer automation. That is a packaging and access change, not evidence that every account has the same configuration or needs the same migration work.

The removal announcement says development for Essential and legacy functionality stops on August 31, 2026 except for critical fixes and breaking-change support. Zendesk plans to remove those features on December 10, 2026. It recommends that affected customers migrate before August 31 to avoid disruption.

The scope matters. The notice says it does not apply to customers who previously bought the Advanced add-on. A separate older Advanced setup guide is itself limited to customers who bought AI Agents Advanced before March 10, 2026. That is why account state, not the blog title or a generic deadline, should drive the next step.

Separate advanced capabilities from legacy migration mappings

Advanced gave teams more configurable conversations, integrations, and reporting. In the current experience, use cases identify customer requests, generative procedures describe policy-guided work, and dialogues provide scripted paths. Review which approach the account uses for a specific job rather than assuming all automation is a fixed flow.

The following table is specifically a migration map for Essential and legacy functionality, not a list of features being removed from every former Advanced account:

This mapping comes from Zendesk’s migration guide. It is useful for an affected legacy configuration, but it does not prove that a former Advanced customer should rebuild an agent. Review the actual account first.

Find the account state before you plan a migration

In Admin Center, go to AI > AI agents > AI agents. Zendesk’s migration guidance says an Essential (legacy) agent has no Intents or Answers tab. A legacy agent has an Answers tab and may have an Intents tab. Those are practical signs that the account is using technology covered by the migration guidance.

For an Essential or legacy agent, Zendesk says you can manually recreate the functionality now. Its in-product migration tool is described as coming later in 2026, so do not plan a rollout around a tool you cannot yet use. Complex legacy answer flows, especially flows with many branches or integrations, need a careful inventory before any rebuild.

The new experience also changes the operating model:

• Essential agents could be omnichannel. A new AI agent is single-channel, so messaging and email need separate agents. Zendesk’s current creation guide covers messaging, email (including API and web form), and voice workflows.
• Standard responses become system replies. The new experience adds reply types with no Essential equivalent, including unsupported-language, failed-escalation, and technical-error replies.
• Agents are managed in the AI agents workspace instead of Admin Center. Custom translations are also organized differently.

Create and test a current AI agent

For a new build, use Zendesk’s current creation guide, not the older Advanced-only instructions. For a messaging agent, Zendesk’s path is: AI agents workspace > Dashboard > Create AI agent > Messaging. Select a brand and knowledge base, add an existing or new web crawler if needed, personalize the agent, review its system replies, then save/close or activate it when the configuration is ready.

That is setup, not validation. Test a bounded customer job and its failure paths before activation:

1. A straightforward request that the agent should resolve from the approved knowledge base.
2. A request with missing information or a source it must not use.
3. An action that is unavailable or fails.
4. An escalation, including the receiving team’s availability and the customer-facing fallback.
5. Each channel the account needs, because a single new agent cannot cover both messaging and email.

Review the resulting conversation, ticket events, customer wording, and receiving queue. A migration is complete when the new configuration handles the intended job and fails safely, not when old labels have been copied into new fields.

Inspect the support job with eesel CLI before expanding it

If you are reviewing a former Advanced deployment, list the jobs you want an AI to own rather than copying a feature label. For an eesel teammate connected to Zendesk, a person or coding agent can use the eesel CLI to inspect the same configuration and workspace as the dashboard.

Consider a team adding an order-change job. Ask Claude Code, Codex, or Cursor to read the teammate's standing instructions and automations, then prepare a proposal that identifies the policy, required lookup, approval point, and human exception route. Review connected permissions and get the workspace owner's authorization before any chat or write.

npx @eesel/cli --agent "Zendesk support" instructions
npx @eesel/cli --agent "Zendesk support" automations
npx @eesel/cli --agent "Zendesk support" chat "Propose a bounded order-change workflow from our approved policy. Identify required information, checks before any change, and cases a human must own. Flag missing knowledge or access. This is a proposal only: do not create an automation, modify an order, or message a customer."

Node.js 18.17+ is required, and every command returns JSON. A script can keep the configuration snapshot with the proposal. The coding agent can flag a conflict, such as instructions permitting an address change while a related automation could send confirmation before review. The owner must verify that interpretation, approve the precise correction, and test an allowed case, an exception, and a failed lookup before enabling the job.

Use the documented command's --help and --dry-run to inspect proposed writes. Chat is real work, and approval queues contain only held actions; a request not to act is not a permission control. This workflow improves the eesel teammate. It is not a native Zendesk configuration export or migration tool.

Keep the customer job bounded

The most useful comparison is not “old Advanced versus new.” It is whether the AI has a defined customer job, approved knowledge and actions, and a reliable human route for exceptions. With eesel, use your terminal or coding agent to inspect that job through the CLI, review a proposed change, and widen the scope only after checking the result.

Try eesel