Your greatest automation opportunity might be where your business is different. Custom Agents close the gap.

One of the clearest pieces of feedback we’ve heard from customers is this: Don’t just give us AI. Give us AI that understands our business.

That makes sense. AI agents are already doing a good job with common customer questions and requests. Zendesk customers are achieving automation rates of 80% or more, and Zendesk AI agents are helping more than 10,000 companies resolve billions of interactions each year.

But the conversation is only part of the resolution.

Behind every customer interaction is a layer of operational work. A refund may require checking order history, loyalty status, previous refund activity, and the company’s latest policy. A warranty claim may require reading a serial number from an image, checking coverage in another system, and deciding whether to approve the claim or send it for review.

AI agents can handle parts of these workflows today. But when the next step depends on a business-specific process, policy, or system, the work often bounces back to a human agent.

This is the messy middle of service. What exactly makes it so messy? Every business has its own data, systems, policies, and processes, meaning the back-office work required to move a request towards resolution looks different at every company.

It’s a gap your frontline AI agents, the ones directly conversing with your customers, should not be wired to solve for. But it’s one Zendesk has closed with Custom Agents.

The work behind the conversation

Prebuilt AI agents are a powerful starting point. They help businesses move quickly and automate common service interactions without having to create everything from scratch. Traditional automation is also useful when a process follows a predictable path. If a ticket contains a certain keyword, apply a tag. If an order meets a fixed condition, issue a refund. But real service work rarely stays on the happy path.

The work that matters is often the work that falls outside the standard path. It’s the refund that depends on customer history, the warranty question that requires image analysis, or the emerging product issue that needs to be detected, linked to an existing bug, and routed to engineering. 

Several challenges make this work difficult to automate reliably:

• Unique processes: Business rules, exceptions, approval thresholds, and decision criteria vary from one organization to the next.

• Integration gaps: The information needed to make a decision may be spread across Zendesk, order systems, CRMs, billing platforms, or other internal tools.

• Governance risks: An agent needs carefully scoped access to knowledge and actions, along with clear permissions, guardrails, and escalation paths.

• Cost and complexity: Connecting systems, maintaining logic, and updating automation as policies change can require significant time and technical effort.

These workflows require more than a fixed sequence of steps. They require context, sophisticated reasoning, business knowledge, the ability to make decisions within defined boundaries, and sometimes the ability to monitor and act proactively. 

Custom Agents bring agentic AI to this work. They can evaluate multiple inputs, reason through business-specific instructions, use approved tools and knowledge, and determine the next best action within defined guardrails.

And they’re already moving the needle. In just seven weeks, Zendesk customers ran more than 1 million Custom Agent executions, with some seeing automated resolution rates rise by up to 10%.

Built for how your business actually works

Every business has policies, processes, and decisions that generic automation can’t capture.

For an agent to do anything useful, three things have to be true at once: it needs the right context, the ability to take action, and governance to operate safely. It needs access to the information that matters, permission to use the right tools, and clear boundaries around what it can do.

Custom Agents turn that business-specific expertise into reusable AI specialists. Each one is designed for a particular job—like evaluating refund eligibility, validating warranty claims, detecting account risk, or monitoring recurring product issues—and can work across conversations, workflows, and background operations.

Rather than asking teams to redesign their processes around a generic bot, Custom Agents fit into the way the business already operates, using the right context, systems, actions, and guardrails for each job.

What’s inside a Custom Agent

A Custom Agent has three core parts:

• Inputs provide the context the agent needs at runtime, such as a ticket ID, customer details, or information passed from another workflow.

• Instructions define the agent’s role, responsibilities, policies, and guardrails. This is where you explain what the agent should do and how it should make decisions.

• Outputs define what the agent returns so another workflow, agent, or human team member can use the result.

The agent’s capabilities come from the tools you connect it to. You can mix and match the tools needed for a particular job:

• Zendesk Actions let the agent update tickets, modify user profiles, manage custom objects, or complete other native platform operations.

• Custom Actions connect the agent to external systems through administrator-defined API endpoints.

• Action Flows connect multiple actions into a visual workflow that can coordinate work across systems.

• External Actions provide ready-to-use connections to platforms such as Slack, Jira, and Salesforce.

• Knowledge Articles ground the agent in verified company information, policies, and procedures. The agent can also interpret images and diagrams contained in that knowledge.

• Other Custom Agents let one specialist call on another when a workflow requires more specific business logic.

This structure gives businesses control over what an agent knows, what it can do, and how it contributes to a broader resolution process. A Warranty Agent, for example, can receive a ticket and image as inputs, use instructions grounded in warranty policy, analyze the image, retrieve product details, call an external system, and return a recommendation or approved action.

That is the difference between configuring a generic AI agent and creating an AI specialist for a specific business process.

The proactive layer of service operations

Most service automation starts when a customer submits a request. But Custom Agents can also help teams get ahead of the work.

Custom Agents can run independently of a customer conversation or ticket event, so teams can use them proactively to monitor service operations on a schedule. 

A Bug Monitoring Agent could scan incoming tickets every four hours, identify a recurring issue, check whether a corresponding bug already exists, create a Jira ticket, and notify engineering in Slack. An Account Risk Agent could review open tickets hourly, identify high-impact issues, and alert the right team before the problem escalates.

This creates a new layer of service operations that is not beholden to the ticket lifecycle. These AI specialists don’t simply wait for work to arrive, but continuously analyze signals, surface risks, and move the organization toward the next action.

Put more expertise behind every resolution

As specialists for your service team, Custom Agents turn your business logic—your policies, decisions, and workflows—into reusable expertise that can be called across Zendesk wherever it’s needed.

AI Agents can call on Custom Agents when a customer-facing interaction requires complex business logic. For example, an AI Agent handling a return request could ask a Refund Exception Agent to evaluate whether an exception is appropriate based on customer history and company policy.

Agent Copilot can use Custom Agents to complete complex work on behalf of a human agent. A Copilot could call a Warranty Agent to analyze an image, extract a serial number, verify coverage, and recommend the next step.

Action Builder can orchestrate the broader workflow, passing the right inputs to a Custom Agent and using its output to trigger the next action.

This creates a connected workforce in which each specialist has a defined role, approved access, and a clear way to contribute to resolution.

Built for production. Designed to improve.

Moving an AI specialist from a demo into real service operations requires more than creating it. Teams need to test how it reasons and acts, understand the sources behind its decisions, control what information and tools it can access, and define how it should escalate when it needs help.

Custom Agents are built into Zendesk, so teams can build, test, deploy, monitor, and improve them in the same environment. Testing and simulation help expose issues before deployment. Traceability shows how an agent reached an outcome and which knowledge or tools it used. Guardrails, permissions, and escalation paths help teams operate it with confidence.

Once an agent is in production, performance and outcome data show where it’s succeeding, where it’s getting stuck, and what needs to change. Those signals fuel the Zendesk Resolution Learning Loop, helping teams continuously improve the agent’s instructions, knowledge, policies, and actions as the business evolves.

Specialization is how service scales

The future of service won’t be powered by one general-purpose agent trying to do everything. It will be powered by a connected workforce of specialists, each built for a specific kind of work and grounded in the context, systems, and policies required to complete it.

Custom Agents let businesses turn their unique expertise into reusable AI specialists that work across conversations, workflows, and background operations. They help teams automate more of the work that makes their business different—while keeping that work connected, governed, and continuously improving.

Specialization is not just an organizational model for AI. It is what makes AI more useful, reusable, and governable. By giving each specialist a defined job, the right context, and controlled access to tools, teams can improve one capability without making a single general-purpose agent more brittle or difficult to manage.

That’s why Zendesk is building toward a connected workforce of specialized agents—each designed for a specific kind of work, but reusable wherever that expertise is needed.