> ## Content Index > Fetch the complete content index at: https://internalnote.com/llms.txt > Use this file to discover other available public pages before exploring further. # Zendesk Specialised AI Agents - URL: https://internalnote.com/zendesk-specialised-ai-agents/ - Published: 2026-09-24T08:08:04.000Z - Updated: 2026-09-24T08:14:57.000Z - Author: Thomas Verschoren - Tags: AI Agents, Custom Agents Zendesk AI Agents are more than just answer bots. They can understand what an end customer is trying to achieve and help carry the request through to a resolution. By combining knowledge, context, procedures and actions, they can interpret a policy, recognise the situation, decide what should happen next and retrieve or update information in Zendesk or a connected system. They can resolve an interaction autonomously, involve a human agent when needed and give your team the insights required to improve the next resolution. That is the promise of AI Agents: a conversational resolution layer that can understand what an end customer needs and act on their behalf. As companies become more familiar with what AI Agents do, two practical problems become clearer. The first is where to start. You may know that AI (Agents) could handle a meaningful part of your business, but opening Zendesk for the first time still presents a platform that needs to be configured. The capability is there; the work is knowing how to get from an empty starting point to a useful set of procedures, knowledge and actions without designing every piece from scratch. The second problem is that not every valuable automation is a conversation. Some work happens beside or in parallel with a conversation, while other work only becomes relevant once the conversation has resolved. A process might inspect an attachment, identify a fraud signal, detect a sales opportunity, update a record in another system or check whether a resolution was complete. Sometimes that work extends a conversation. Sometimes it runs in parallel. Sometimes it should happen without the end customer seeing it at all. 💡 Screenshots in this article are for illustrative purposes only and may not reflect the final product. Features and capabilities are subject to change. # Specialized Agents Specialized Agents are Zendesk’s answer to these two problems. Industry Agents provide built-in workflows for jobs that are common within a particular industry. Custom Agents sit at the other end of the spectrum, adding focused logic for work that is specific to your organisation. > “Industry agents for work that's common for your industry, and custom agents for the work that's specific to you.” Zendesk’s new Industry Agents shift the platform away from a set of vertical capabilities such as AI Agents, Copilot and Action Flows, towards a horizontal solution that solves a use case across the platform. Instead of asking you to decide which feature to configure first, they start with the work you want to get done and assemble the relevant pieces around it. That distinction is important because an Industry Agent is not a new kind of AI Agent or a separate runtime. It is a pre-configured package of workflows assembled from the existing Zendesk platform capabilities. Depending on the use case, the package can include AI Agent procedures, Agent Copilot procedures, knowledge, Action Flows, Custom Agents, Custom Objects, connectors and the standard configuration needed to make the process work. ![](https://storage.ghost.io/c/ca/c0/cac0c82b-bc4d-4404-9eb4-9cc75e8d045e/content/images/2026/09/Platform-Advantage-3.png) Launching an Industry Agent does not give you a different Zendesk or a different kind of AI. It gives you a more useful first configuration: one that starts with the context, procedures and actions associated with a particular industry job. Once it is running, the underlying objects remain part of the normal Admin Center experience. You can inspect, test, edit and extend them in the same way as other AI Agent, Copilot, procedure, knowledge, Custom Agent and Action Flow configuration. # Industry Agents E-commerce is the first industry where this approach becomes concrete. E-commerce Industry Agents give you a series of built-in workflows that already understand the jobs a retailer needs to perform. These include workflows for returns, refunds and exchanges, order management, order tracking and status, promotions and discounts, fulfilment and delivery, billing and payments, and missing or incorrect items. ![](https://storage.ghost.io/c/ca/c0/cac0c82b-bc4d-4404-9eb4-9cc75e8d045e/content/images/2026/09/Detail-1.png) **The Launchpad design presents built-in retail workflows as selectable AI Workflows, with the process steps visible before publication.* Take a refund. An end customer may ask a simple question: “Can I get my money back?” Resolving it may require Zendesk to identify the person, retrieve the order, understand when it was placed, check the payment method, apply the refund policy, assess risk, confirm the appropriate amount, issue the refund and explain what happens next. An AI Agent can be configured to do that work. But it takes time to configure and add all that context. A pre-configured e-commerce use case starts with the job already understood. It provides the procedure for handling the interaction, the knowledge needed to explain the policy, the actions needed to retrieve and update information, and the Action Flows that connect those steps. Where the process needs flexible reasoning, the configuration can include a Custom Agent for that focused responsibility. The connected systems behind these workflows can include Shopify, WooCommerce, Stripe, Riskified or any other connected system, depending on your configuration and the job being automated. The platform remains the place where the process is composed, governed and improved, even when the work reaches into those external systems. ## Setup The last few months Zendesk has been focused on adding Launchpads to most of its products. We’ve seen it when launching Agent Copilot, it’s part of the new Intelligent Triage experience, and now, it’s also how you deploy your first Industry Agent. ![](https://storage.ghost.io/c/ca/c0/cac0c82b-bc4d-4404-9eb4-9cc75e8d045e/content/images/2026/09/Setup-3-1.png) ![](https://storage.ghost.io/c/ca/c0/cac0c82b-bc4d-4404-9eb4-9cc75e8d045e/content/images/2026/09/Setup-4-1.png) ![](https://storage.ghost.io/c/ca/c0/cac0c82b-bc4d-4404-9eb4-9cc75e8d045e/content/images/2026/09/Setup-5.png) Zendesk detects your industry based on your website and context. It presents the relevant out-of-the-box workflows for common use cases. You can then add your own documentation to provide the business nuance. The system then rebuilds each selected workflow around that context, drafts the procedures and prepares the configuration for publication. In practical terms, the flow is: 1. Zendesk detect your industry. 2. It present the out of the box workflows for popular use cases. 3. You add your documentation and business context. 4. Zendesk uses that context to enrich the knowledge, procedures, Action Flows and Custom Agent components that the selected workflows need. 5. You can optionally add additional custom agent to the process to handle more complex logic. 6. You then review, customise and publish. ![](https://storage.ghost.io/c/ca/c0/cac0c82b-bc4d-4404-9eb4-9cc75e8d045e/content/images/2026/09/Detail-5.png) ![](https://storage.ghost.io/c/ca/c0/cac0c82b-bc4d-4404-9eb4-9cc75e8d045e/content/images/2026/09/Detail-4.png) The wizard is not a black box that leaves you with a new product to administer. Once deployed, you can see the resulting objects in the existing Admin Center surfaces. You can change the procedure, adjust the prompts, add or replace knowledge, edit an Action Flow, change routing or thresholds, add a connector and adapt the Custom Agent instructions, just as you would with any other Zendesk configuration. That matters over time. The wizard gets you to a useful first version, but the resulting configuration becomes part of your normal operating model. A procedure can be reviewed and refined. Knowledge can be updated. An Action Flow can be tested and activated. A Custom Agent can be edited and tested. Insights can show where the workflow needs improvement, so the next version is based on what actually happened rather than what the team expected to happen. # Custom Agents Conversational resolution is the centre of the platform, but it is not the boundary of the platform. An end customer might attach a photograph of a damaged product. A support organisation may want to inspect that image, compare it with a policy and recommend the right next action. A fraud team may want to look for signals across the conversation, the order and the customer record. A sales team may want to identify an opportunity in a support conversation and pass it to the right process. An operations team may want to check every solved ticket for an unanswered question or missed promise. None of these jobs drive the conversation. They are specialised capabilities that can sit beside a conversation, support it or run after it. That is where Custom Agents fit. They let you define a focused piece of business logic with custom instructions, approved knowledge and the actions it needs. They can reason over the context they are given, interpret information that does not fit neatly into a deterministic rule and return a defined result to the workflow that called them. An OCR-style Custom Agent might inspect an attachment and extract the information a downstream process needs. A fraud Custom Agent might evaluate signals and return a risk assessment. A lead-detection Custom Agent might read the conversation and identify whether a sales opportunity is present. A resolution-audit Custom Agent might review a solved interaction and flag an unanswered question before the ticket is closed. ![](https://storage.ghost.io/c/ca/c0/cac0c82b-bc4d-4404-9eb4-9cc75e8d045e/content/images/2026/09/Frame-1.png) The important point is that these Custom Agents are focused. They are not intended to replace the conversational AI Agent or become a second customer-facing front door. They specialise in one responsibility that can be reused wherever that responsibility is needed. This is also why Custom Agents do not need to “talk” to the end customer in the same way as an AI Agent. They can return the result of a focused task to the surrounding workflow, which then decides what to do with it. The conversational layer can ask for missing information or explain the outcome; the specialised agent can concentrate on the job it was created to perform. DIVE INTO CUSTOM AGENTS For more information on the role of Custom Agents, read: - [Testing Zendesk’s new Custom Agents](https://internalnote.com/testing-zendesk-custom-agents/) - [Closing the Resolution Loop with Custom Agents](https://internalnote.com/closing-the-resolution-loop/) - [Action Platform. The new way of automating Zendesk](https://internalnote.com/automating-with-action-platform/) # Intelligent Service Platform Zendesk has always had the building blocks needed to create powerful automations and resolve conversations at scale. Industry Agents use that foundation to help you get more from the platform. ![](https://storage.ghost.io/c/ca/c0/cac0c82b-bc4d-4404-9eb4-9cc75e8d045e/content/images/2026/09/Act-1.png) The *AI Agent* still owns the interaction. It understands what the end customer is asking, selects the appropriate procedure, retrieves relevant knowledge, asks for missing information, takes action and explains the result. *Agent Copilot* gives a human agent assistance with similar work while the person remains responsible for the conversation. *Procedures* describe how the interaction should progress. They provide the decision path, the questions to ask, the conditions to check and the outcome to reach. Knowledge provides the information that grounds the answer. Actions perform the work against Zendesk or another system. With built-in use cases and procedures, you start with an informed first draft rather than a blank page. *Action Flows* connect actions into a deterministic process. They are useful when the system should follow defined steps, branch on known conditions, repeat a task, call an external service or return a structured output. They turn several actions into a reusable workflow. *Custom Agents* are useful when the work needs runtime reasoning. They can interpret information, decide how to apply a set of instructions and use approved tools to reach an outcome. Their path can adapt to what they discover, while the surrounding Action Flow or procedure still defines where they fit and what should happen with their result. ## The platform Industry Agents change the starting point for AI in Zendesk. Instead of beginning with a vertical list of capabilities and asking how they might fit together, you can begin with a use case and see the combination of capabilities required to resolve it. > “The era of one-size-fits-all AI Agents is over.” That does not mean every organisation needs a collection of disconnected agents. It means the platform should apply the right combination of conversation, assistance, deterministic automation, specialised reasoning and improvement to the job at hand. [AI has grown up. Has your service model kept up?AI isn’t just answering questions anymore. It’s taking action, reshaping teams, and changing how businesses operate.![](https://storage.ghost.io/c/ca/c0/cac0c82b-bc4d-4404-9eb4-9cc75e8d045e/content/images/icon/al2o9zrvru7aqj8e1x2rzsrca-a2bca463-5f0f-4c62-917d-60d96c0ae0cf.ico)LinkedInZendesk![](https://storage.ghost.io/c/ca/c0/cac0c82b-bc4d-4404-9eb4-9cc75e8d045e/content/images/thumbnail/1787159400695-7f2a2655-e361-4a69-ac1a-e1214c527f21.png)](https://www.linkedin.com/pulse/ai-has-grown-up-your-service-model-kept-zendesk-lc6pc/?ref=internalnote.com) Industry Agents provide a faster starting point for an industry job. Custom Agents extend the platform wherever the built-in workflows or ordinary conversation do not cover your specific needs. Action Flows can run defined work in the background, while [Admin Copilot](https://internalnote.com/introducing-admin-copilot/) and its proactive insight layer help the people responsible for Zendesk understand what happened and improve the configuration behind it. These are not permanent, competing operating models. A single use case can move between them. An AI Agent can handle the conversation, an Action Flow can update an external system, a Custom Agent can interpret an attachment, Agent Copilot can support a human hand-off and reporting can show where the process needs improvement. Industry Agents make that combination easier to start by organising it around the work rather than the product menu. ## The future The future of Zendesk is not a monolithic agent that tries to own every kind of work. It is an Intelligent Service Platform that can specialise around the industry and the use case while continuing to build on the same shared foundation of knowledge, context, procedures, actions and insights. Industry Agents make the platform horizontal: a business problem can draw on AI Agents, Copilot, Action Flows, Custom Agents and the rest of Zendesk wherever they are useful. The result is a quicker path from the blank page to a working resolution, with the control to customise and improve the underlying configuration as your operation evolves. ## Sign up for Internal Note Turning Zendesk into practice. – A weekly newsletter written by Thomas Verschoren. Subscribe Email sent! Check your inbox to complete your signup. No spam. Unsubscribe anytime.