Every Zendesk Relate keynote tells a story. Two years ago, we introduced AI for customer service. Last year, we sharpened that to a single word: resolution. This year, we take the next step: the Autonomous Service Workforce. AI that works deeply in your service operations.
The framing is simple. The Resolution Platform now runs as two connected loops. One handles the customer’s interaction. The other makes the platform better at handling the next one. Specialized AI agents handle the resolution work. Proactive copilots keep the system improving. The connected platform feeds both. Together, they form a workforce that moves on its own.
That is the vision. But a keynote vision is not the same as something you can just switch on. The real question after any product keynote is: what can I actually use today?
This article answers that question for Relate 2026. Five releases you can get your hands on right now. Some are generally available, ready to turn on and put in front of customers today. Others are in early access, which means you can start testing them in your own instance now rather than waiting for a general release.
1. Omnichannel agentic AI agents, now the default
For most of their history, Zendesk’s AI agents worked in two ways: they followed a hand-built flow, or they generated answers from your knowledge.
The flow approach was the older one. You build a decision tree by hand: ask for an order number, check if it’s valid, branch to a refund path or an exchange path. Every step or exception had to be defined in the builder. That made flows rigid to use and complex to maintain.
Agentic AI changes the model. Instead of building a flow, you describe a procedure. You write down the process in plain language. The AI agent reads that procedure and works out how to execute it.
The key word is adaptive. The agent does not march through your procedure like a script. It reasons across it. If the customer provides all the context in their first message, the agent skips the questions it no longer needs. If the customer cannot find their order number, the agent helps them locate it, then picks up where it left off. It loops, adapts and recovers, while staying inside the boundaries you set.
Agentic AI is now the default across all our channels: email, messaging and voice. The same reasoning that handled conversations now reads incoming emails, answers from your connected knowledge, and runs business procedures or accesses connected integrations. No hand-built flow required, on either channel. Voice is the third channel in the picture, with agentic voice AI in early access now.
The impact of going agentic
- Speed of deployment. A written procedure is faster to stand up than a flow.
- Ownership shift. A flow builder rewards technical skill. A procedure rewards process knowledge. If you know how your team handles a refund, you can write the procedure. This moves AI setup closer to the people who actually own the process.
- Maintenance. Changing a flow means rebuilding parts of the chart. Changing a procedure means changing words. As your processes evolve, the difference compounds.
There is a catch worth stating plainly: agentic agents are only as good as the process you describe. If you don’t know how you handle a refund, the agent won’t figure it out for you. So before you build, define and understand how you actually want to work.
2. Custom AI agents through Agent Builder
A single capable AI agent is useful. A team of specialized agents is a workforce. Agent Builder is how you build that team, and they open up a future of specialized agents for specific needs.
Most AI agents today are general-purpose and they handle a use-case end to end. That works for a lot of cases. But real operations are not one undifferentiated stream of questions, they are a combination of different processes and decisions.
Agent Builder lets you create custom AI agents tuned to specific workflows, using natural language to build, test, deploy and improve them. Rather than one agent trying to be everything, you build specialized agents that each know their domain. They can run as part of a customer interaction, or alongside it running backoffice processes.
Paired with this are action flows for AI agents. They now use the same building blocks already used in other Zendesk automations like Agent Copilot or Action flows and they let agents act: create a Jira issue, look up an order in Shopify, post to Slack, update Salesforce, or call a custom API.
The headline impact is simple.
Your AI agents can complete work, not just answer questions. An agent that can look up an order, process a return, update the ticket and notify the warehouse is doing the job, not deflecting it.
Generalization improves quality. A custom agent tuned to one workflow can be far more effective than a generalist trying to cover everything. And reuse matters: the same action can serve an AI agent, an Agent Copilot procedure, or a ticket-triggered flow. Build the integration once and use it across the platform.
3. Admin Copilot
Most AI attention goes to the front line. Admin Copilot points it at your Zendesk setup itself.
It gathers insights from across your instance and gives you an overview in Admin Center of what’s occuring. More than that, it recommends improvements: modify a view, change a routing rule, add a trigger. The recommendations come from real ticket data, so they react to what is actually happening.
Admin Copilot is also conversational. You can investigate how something works, describe a change you want to make, refine what it proposes, and deploy from inside that conversation.
Powering your learning loop
Turning customer processes into triggers, views and automations has always been specialist work. Admin Copilot gives that capability to your own team leads.
Admin Copilots are also proactive, changing the speed of optimization. Instead of reviewing analytics once a quarter and scheduling configuration work later, you now get notified of an issue, see steps to fix it, and adjust. Configuration becomes a continuous practice.
It also helps with diagnosis. When rules behave strangely, finding out why has always been a treasure hunt. Admin Copilot can surface the trigger, rule or routing issue behind the symptom..
4. IT asset management
Zendesk has been widening from customer service into employee service, allowing you to offer the same quality of service you give to your customers to your own team. Approvals, Tasks and a Service Catalog, and a list of native integrations with tools like Workday, Intuo or JAMF already helped IT and HR run their operations in Zendesk. IT asset management is the next piece.
ITAM gives you a unified place to track, monitor and manage technology assets inside Zendesk Support: laptops, phones, monitors, software licences, whatever your organisation issues and supports. The asset details integrate directly into ticket workflows, so when an employee raises a ticket about their device, the relevant asset information is already there.
Context makes for better resolutions.
Employee service tickets are often about a thing: a device, a licence, a piece of equipment. With ITAM, the agent has access to the asset data that is already part of the ticket. And when changes are made, those changes reflect on the assets and into connected systems
5. Knowledge Connectors
Every resolution depends on knowledge. The problem is that knowledge often doesn’t live in one place.
Help Center holds support articles. But the real operational knowledge of a company is scattered: runbooks in Confluence, policies in SharePoint, specs in Google Drive, project notes in Notion. Knowledge Connectors bring that scattered knowledge into Zendesk’s knowledge graph without forcing anyone to move it.
The content stays where it lives. The connector indexes it and makes it searchable and available across the platform: Help Center search, generative search, AI agents and Agent Workspace. The set of supported sources keeps growing.
Knowledge is at the foundation of resolutions.
The first impact is reach. Every silo you connect is another body of knowledge the agent can draw on.
The second is consistency. A unified knowledge graph means your AI draws from one harmonised foundation instead of fragmented sources.
The third is that native connectors fit the platform better than custom indexing. Connected content appears in Agent Workspace, and usage feeds back into analytics and improvement suggestions. That makes knowledge part of the learning loop, not a static index.
Pulling it together
Five releases. All within reach today.
Taken together, they make the Autonomous Service Workforce feel less like keynote language and more like a system you can actually build with. Agentic AI agents give you the front line. Agent Builder lets you specialize. Admin Copilot improves the setup. ITAM and Knowledge Connectors expand the systems underneath.
The thread running through all of them is simple: Zendesk is no longer a ticket platform with an AI agent in front. It’s a resolution platform that reasons on your processes. It uses every interaction to improve the next one and helps you deliver actual resolutions, not just answers, to your customers and employees.
The platform has moved a long way toward doing the work. But it still rewards teams who know how they want to work and can say so clearly. That has always been the real skill in customer service. It is even more so now.
