Customer service automation has made routine requests easier to resolve. Yet many service teams still spend hours coordinating the work behind more complicated issues.
Think of a customer reporting a defective product. Resolving the issue may require checking a warranty, reviewing purchase history, updating internal systems, and arranging a replacement. Each step adds work between the customer's initial request and the final resolution.
Custom AI agents are changing how businesses handle these processes. By connecting business knowledge, policies, and operational systems, they automate workflows specific to each organization.
The shift is already producing measurable results. According to an Anthropic customer story, Zendesk Custom Agents reached one million executions in their first seven weeks of early access.
Customers running these agents have also reported up to 10 percent higher automated resolution and up to 80 percent lower human handle times. These early results point to an opportunity for CX leaders: Extend AI automation into the complex operational work behind successful service resolutions.
Key takeaways
Zendesk Custom Agents reached one million executions in seven weeks of early access, demonstrating early adoption of specialized AI workflows.
Custom AI agents automate business-specific service tasks that require context, defined processes, and actions across multiple systems.
Customers reported up to 10% higher automated resolution and up to 80% lower human handle times when running Custom Agents.
Trusted knowledge and controlled system access support reliable automation, with permissions and oversight governing what agents do.
Human agents gain more capacity for complex customer needs as AI handles repetitive coordination and administrative work.
What is complex customer service workflow automation?
Complex customer service workflow automation uses AI and connected business systems to execute multi-step processes involved in resolving customer requests. These processes often involve information from several applications, company-specific policies, and coordinated actions across teams. For example, a warranty claim may require an agent to verify purchase details, review coverage, check replacement inventory, and initiate an approved action.
Traditional automation works well for predictable tasks with predefined steps. More complicated requests involve exceptions, changing information, and business-specific decisions. Custom AI agents extend automation into these workflows.
Unlike customer-facing AI agents that manage service conversations, specialized custom agents focus on operational tasks behind those interactions. They use approved knowledge, business instructions, and connected tools to complete specific processes and escalate cases when human intervention is required.
Why complex service workflows remain manual
Many service requests look simple at first. A customer wants a refund. Someone needs an account issue resolved. A business receives hundreds of tickets reporting the same technical problem.
The complexity often emerges after the request reaches the service team. Employees must gather information from different systems, interpret company policies, coordinate with other departments, and update records. These steps create delays and consume resources.
Businesses have traditionally addressed this work through manual processes or custom-built automation. Larger organizations may have engineering teams capable of developing specialized workflows. Smaller teams often rely on employees to handle requests individually. Both approaches present challenges as service demand grows.
Building one-off automation requires technical resources and ongoing maintenance. Manual coordination increases handling time and creates opportunities for inconsistent outcomes. The result is a gap between responding to a customer and completing the work needed to resolve their issue.
How Zendesk Custom Agents reached one million executions
Zendesk developed Custom Agents to give businesses a way to automate workflows unique to their operations. Powered by Anthropic's Claude, these agents follow natural-language instructions and connect to business systems through approved tools.
According to Anthropic's Zendesk customer story, a five-person Zendesk team built the Custom Agent Builder using Claude Code and the Claude Agent SDK. The team progressed from proof of concept to early access in four months. During the first seven weeks of early access, Zendesk Custom Agents completed one million executions.
Research note: An execution represents an agent run, not necessarily a completed customer service resolution. The milestone demonstrates early usage, while automated resolution and handle-time metrics measure separate outcomes.
The technology also runs through Amazon Bedrock within Zendesk's existing Amazon Web Services (AWS) environment. This approach supports data locality and governance requirements by keeping model processing within Zendesk's established AWS hosting regions.
For businesses adopting specialized agents, these controls matter as much as the ability to automate. Organizations need reliable outcomes, visibility into agent activity, and clear boundaries around the actions AI performs.
How custom AI agents handle complex service work
Custom agents bring together information, instructions, and connected tools to complete defined tasks. Their value becomes clearer when examining the operational work surrounding a customer request. Here's how custom AI agents handle complex service work.
Gather information from connected systems
An agent may need customer history, purchase records, service policies, or information stored in external applications. Access to this context reduces manual searching and gives the agent the information required to complete its assigned task. With connected knowledge, teams also reduce reliance on disconnected information sources.
Follow business-specific processes
Each business has different rules governing refunds, warranties, account changes, and service exceptions. Custom agents use instructions written in natural language to follow these processes.
Businesses define the agent's responsibilities, available tools, and permitted actions. These boundaries establish how the agent approaches its work and when it should escalate a request.
Coordinate work across applications
Complex requests frequently involve multiple systems and departments. A technical incident, for example, may require engineering involvement, customer notifications, and ticket updates.
Anthropic describes a Zendesk Custom Agent that identifies related incident tickets, creates a Jira issue, and notifies engineers through Slack. The agent also updates affected customers. Once the issue is resolved, it closes related tickets and records an explanation.
This connected workflow reduces manual coordination and keeps the resolution process moving.
Preserve human oversight
More sophisticated automation requires clear controls. Organizations need to know which actions their agents execute, what information they access, and when human review is necessary.
Zendesk Custom Agents support defined action permissions. Zendesk is also developing monitoring and testing capabilities, including tool-call logs and simulated conversations. These controls enable teams to evaluate performance, investigate unexpected behavior, and refine their automated workflows.
What results are businesses seeing?
The first seven weeks of Zendesk Custom Agents' early access provide an indication of their potential.
Anthropic reports that customers running Custom Agents have achieved:
Up to 10% higher automated resolution: More customer requests reach a resolution through automation.
Up to 80% lower human handle times: Employees spend less time completing work associated with service requests.
The customer story also describes an organization handling more than 14 million service requests annually. During a pilot, this customer used a Custom Agent to automate duplicate ticket merging with 99 percent accuracy.
Duplicate tickets create additional work, distort workload reporting, and require manual cleanup. Automating their identification and consolidation reduces administrative effort at scale.
These examples reflect two different forms of value. Higher automated resolution means more requests are completed without human intervention. Lower handle times mean employees spend less time on cases that still require their involvement. Both contribute to greater service capacity.
The results also reinforce the distinction between ticket deflection vs. resolution. A workflow may execute several tasks without completing the customer's request. Businesses need to evaluate the final outcome alongside the work performed.
How to expand AI automation with confidence
Custom AI agents introduce opportunities to automate service processes that were previously difficult to manage without human involvement. Successful adoption requires clear responsibilities, reliable information, and ongoing evaluation. Let's explore strategies to expand AI automation with confidence.
Start with a specific workflow
Identify recurring tasks that consume significant employee time. These can be incident triage, duplicate ticket management, warranty processing, and internal service requests are potential starting points. Prioritize work with defined objectives and measurable outcomes.
Connect the right systems and knowledge
Agents need access to information relevant to their assigned responsibilities. Connect approved data sources, establish clear instructions, and define which actions the agent may perform. Reliable context and appropriate permissions reduce errors and support consistent execution.
Test and monitor outcomes
Evaluate agents against common scenarios, unusual cases, and conditions requiring human intervention. Review action logs, workflow completion, resolution quality, and escalation patterns.
As automation expands, maintain visibility into what agents do and where their performance needs improvement. These practices align automation with the broader goal of customer service: accurate, consistent resolutions that businesses and customers trust.
Customer service extends far beyond the conversation between a customer and a support agent. Behind each resolution are processes, decisions, and actions that often involve multiple systems.
Custom AI agents bring automation into this work. Zendesk's first million Custom Agent executions and early customer results illustrate the growing demand for specialized automation.
As businesses extend AI into complex workflows, human teams gain more capacity for the problems requiring judgment, empathy, and expertise. The opportunity is to resolve more customer issues while maintaining the control, visibility, and service quality those outcomes require.
Automate complex service workflows with confidence
Reduce manual work, connect business systems, and accelerate resolutions with Zendesk AI agents. Give your team more capacity for complex customer needs while maintaining visibility and control. Start your free trial today.
Frequently asked questions
Custom AI agents are specialized agents that automate business-specific workflows using approved knowledge, instructions, and connected systems. They perform defined tasks such as processing requests, coordinating actions, and updating records.
Custom AI agents automate operational tasks that require information from multiple systems and actions based on business rules. They reduce manual coordination, expand automated resolution, and shorten handling time.
Common examples include incident triage, duplicate ticket management, warranty claims, refund processing, and account administration. The appropriate workflows depend on business requirements, connected systems, and governance controls.
Businesses should track automated resolution rates, workflow accuracy, human handle time, escalation rates, and successful task completion. These customer service metrics distinguish meaningful outcomes from agent executions or activity alone.
Courtney Burry
Vice President of Product Marketing at Zendesk
Courtney is the Vice President of Product Marketing at Zendesk. She has worked in product and technical marketing roles for close to 30 years across a range of startups and Fortune 500 companies including National Semiconductor, VMware and Collibra. Prior to joining Zendesk, she was at Amplitude and spent three years leading product marketing, customer marketing and partner marketing. She loves building businesses and developing teams. She is also an avid vibe-coder and is incredibly passionate about AI.
Originally from Canada, Courtney is currently living in the Bay Area. She has also spent time living and working in Japan, Holland, the Philippines and New York.
Put Custom Agents to work on your complex workflows
Zendesk Custom Agents automate that work across connected systems, reducing manual effort while maintaining visibility and control.
