We Played with Custom Agents and… We Enjoyed It!

By Benoit Smagghe, Premium Plus

As announced in this year's Zendesk Relate, Zendesk is turning to an Autonomous Service Workforce, and one of the biggest announcements was the release of Custom Agents.

The Early Access Program was issued just after Relate, and we naturally enrolled. And you know what? IT.BLEW.OUR.MIND.

With Custom Agents, Zendesk allows your teams to automate and have clear governance on your processes.

But… What is a Custom Agent?

A Zendesk Custom Agent is a task-oriented AI agent that follows a defined procedure and can work on its own or be reused across multiple workflows, including Action Flows, AI Agent procedures, and Copilot procedures.

Custom Agents can be fed with:

• Inputs and outputs: where inputs are the source that starts the procedure (e.g. ticket ID, user information) and outputs are the data the Custom Agent passes over to a ticket, another Custom Agent, an Action Flow, etc. 
• Articles: for the Custom Agent to read a policy and adapt its reasoning.
• Actions: that can be Zendesk's, external, or custom ones.
• Action flows.

As they leverage agentic reasoning, creating procedures is as simple as telling a human agent how to get things done. The procedures should be task-oriented and broken down into a step-by-step guide.

But… What Can You Do with Custom Agents?

As told above, Custom Agents are task-oriented and can be reused in a lot of Zendesk workflows, so you can create a network of Custom Agents that will take care of automating parts of your processes.

So you'll be able to create:

• Order Cancellation Agent
• Returns Agent
• Billing Agent
• …

These agents can work together in combination with automated workflows, and with human and AI agent procedures.

When to Use Custom Agents

Custom Agents are the right fit when a specific task needs to be reused across different workflows, so that editing one process means editing a minimum number of Custom Agents rather than rebuilding logic in multiple places.

Another sign a Custom Agent is the right choice: simple integrations, where the Custom Agent runs before a human agent's action or interaction, since it may need time to reason, interact with your stack, or deliver its output.

Some Practical Use Cases

We like Custom Agents so much that we add them to our demo flows. Some clients come with specific use cases and, in each and every discovery call, we detect the opportunity to build specific Custom Agents that will meet their needs.

We have created specific Custom Agents for:

CX use cases

We mainly focused on extracting information from documents provided by requesters, and we have built:

• Invoice Manager: looks into provided invoices and extracts the order number, invoice number, invoice amount, and order date 
• ID Reader Agent: extracts the name, ID of the document, and birthdate
• Medical Document Agent: reads a medical document and extracts the doctor's name, patient's name, and certificate validity

The flow is largely the same for these agents: extracting relevant information and storing it in dedicated fields speeds up processes for checking authenticity against a third-party database, verifying required documents are complete, automatically enriching client information for escalation, and passing that information to other custom or human agents.

EX use cases

• Problem Agent Handler: detects any incident that can automatically be linked to an existing problem. When the problem ticket is updated, all linked incidents receive a notification about the progress of their request. 
• Loan Asset Manager: when a requester's asset loan is approved, this Custom Agent searches the asset database for an available asset, assigns it automatically, sets the pickup date according to loan policy, and informs the requester that the asset is ready for pickup. 

CX and EX use cases

Duplicate Handler Agent: automatically merges similar tickets from the same requester and adds an explanatory internal note about the merge, work that agents currently have to do manually by reviewing the merge suggestions panel themselves.

To Go Further 

Custom Agents bring real possibilities, but also new risks if they aren't set up properly, scalability, a governance framework, and alignment with the rest of your Zendesk instance need to be built in from the very beginning, not treated as an afterthought.

At Premium Plus, we align the use of AI with your business strategy while maintaining a high level of governance, keeping processes consistent and AI costs under control.