Zendesk's AI Agent Launchpad Writes Drafts From Your Ticket History. Here's What You're Approving.

On 5 November 2026, Zendesk started rolling out AI agent launchpad, a guided setup that reads your ticket history and proposes the help center articles and procedures your AI agent should run on. It is a sensible answer to a real chore. It also changes what an admin is approving when they click "create", and that is worth thinking through before you switch it on.

What launchpad does

Zendesk's announcement describes three parts. For onboarding: "Instead of starting with a blank configuration, launchpad analyzes your account ticket history to detect common customer topics. It automatically generates and proposes procedures and help center articles, then tests your AI agent against high-impact scenarios to validate readiness before go-live."

After launch, recommendations keep coming: "AI agents continuously evaluates your ticket trends to surface suggestions for new generative procedures and help center articles. Admins can generate content in one click, save recommendations for later, or skip suggestions with contextual feedback to refine future suggestions." Zendesk's recommendations article adds that they "update every week, and change as your account data changes."

A new overview page for each AI agent shows "key metrics including automated resolutions (AR), total conversation volume, and understood conversations."

Zendesk's reason: "Deploying and tuning an AI agent traditionally requires substantial manual effort, from identifying user intent trends to authoring articles and testing workflows."

It is not for every Zendesk account. The announcement says: "AI agent launchpad is available to customers on the latest AI agent infrastructure." The rollout runs from 5 to 30 November.

What stays in the admin's hands

Nothing here publishes on its own. Zendesk's recommendations article is specific. When an admin creates an article from a recommendation, "the article opens as a draft in the article editor where you can review, edit, and publish, if you want." A procedure "opens for you to review and edit", and goes live only when the admin clicks Publish procedure. Each recommendation also shows an "Automation potential" estimate and a "Rationale", so the admin can see why it was suggested.

So the review step exists, and it is the admin's. The question is what they are reviewing.

Your ticket history becomes the source

Zendesk says launchpad works by "grounding your agent configuration in actual ticket data." That is the strength of the design, because ticket history is where customers' real questions are. It also means the draft articles start from how your team has answered those questions so far.

Ticket history is a record of what happened, not of what should happen. In most support queues it also contains:

• Workarounds that agents used while a bug was open, and that no longer apply.
• Old answers about prices, policies or features that have since changed.
• Exceptions, such as a refund granted once as goodwill, that were never meant to become policy.
• Inconsistency, where different agents answered the same question differently.

A draft built from that history can sound authoritative while repeating any of it. Reviewing it means checking it against current policy, not just reading it for tone. That takes longer than "one click", and it is the part that should not be skipped.

What to check before you approve

• Is the answer still true? Check each draft against your current pricing, policies and product, not against what agents used to say.
• Is it a rule or an exception? If a draft says customers can get something, make sure that is the policy and not a one-off.
• Who owns the topic? Have someone who knows the policy, not only the AI agent admin, read drafts on refunds, billing or security.
• Clean the history first. If you know a topic was handled badly for a while, write that article yourself before the recommendation does.
• Read the metric carefully. In Zendesk's overview article, "understood conversations" includes conversations that were "directly escalated to a human agent without an answer being provided". We wrote about how resolution numbers get counted in what counts as a resolution.

How DeskLeap does it

DeskLeap takes a different route, and it has its own trade-off. We checked our code on 6 November 2026:

• Articles come from your website, not your tickets. "Build from website" reads your site and writes knowledge base articles from it. Your website is something you wrote on purpose. It can also go out of date, so the same rule applies: what the AI learned from is the ceiling.
• Our default is the opposite of Zendesk's. Unless an admin turns on Hold automatically written articles for approval, articles written this way go live straight away. With the setting on, new articles wait in a review queue, and rewrites of live articles are held until someone approves them. If you use DeskLeap and want a review step, turn that on.
• Gaps become a to-do list. Suggested questions come from two places: gaps the AI spots while reading your website, and questions customers actually asked that the AI Agent couldn't answer before it handed the chat or ticket to a person. An owner writes the answer, and that answer becomes the article. A person writes these, so the person is the review.

Neither design removes the need for judgement. Zendesk's launchpad gives the admin a draft drawn from real tickets. Ours writes from your website and, by default, trusts it. Either way, the AI can only be as good as what it learned from, and that is worth checking before customers read it.

Separately, Zendesk also announced on 5 November that Gainsight, eGain, Intercom, Skilljar and Superhuman Docs can now be connected as knowledge sources, so a team can point Zendesk at content it already keeps in those tools. For Zendesk's AI agent pricing, see our earlier post.

DeskLeap is a ticketing, live chat and knowledge base platform. The platform is 100% free with unlimited human agents; an optional AI Agent add-on is $29/mo each.

Sources

Zendesk, Announcing AI agent launchpad (rollout 5–30 November 2026; edited 5 November 2026; read 6 November 2026). Zendesk, Using tailored recommendations to automate manual tasks and optimize workflows in AI agents and Using the agent overview to monitor and manage a specific AI agent (both updated 5 November 2026; read 6 November 2026). Zendesk, Announcing Gainsight, eGain, Intercom, Skilljar, and Superhuman Docs as external content sources (rollout 5–12 November 2026; read 6 November 2026). Quotations are verbatim from those pages. DeskLeap's behaviour is described from our own code as of 6 November 2026.