Sponsored by Zendesk
Agentic AI is revolutionizing three pivotal domains of work: creative tasks, software development, and customer support, according to Shashi Upadhyay, Zendesk’s president of engineering, AI, and product. Among these, customer support stands out as uniquely challenging.
“Customer support is distinct because it places an autonomous AI directly in front of your clients,” Upadhyay explains. “You must trust that the AI will act in the best interest of both the customer and the business. Every advancement in AI should enhance reliability and trustworthiness for customers and human agents alike.”
Zendesk, recognized recently as a Leader in the CRM Customer Engagement Center category, began deploying AI agents around 18 months ago. Their data shows these AI agents independently resolve nearly 80% of incoming customer inquiries. For the remaining 20%, the AI seamlessly escalates the issue to a human agent to handle more complex cases.
“AI agents operate nonstop, 24/7, eliminating wait times and queues. When a customer has a question, the AI delivers an immediate response,” Upadhyay notes. “This leads to higher resolution rates, increased automation, and improved customer satisfaction scores. Given the strong performance of AI agents, we anticipate widespread adoption of this technology across industries in the near future.”
To elevate usability, deepen insights, and accelerate value delivery for businesses of all sizes, Zendesk continuously tests and integrates cutting-edge models like ChatGPT-5. Additionally, the acquisition of HyperArc, an AI-native analytics platform, has significantly enhanced Zendesk’s real-time generative AI insights and analytics capabilities.
Building and Refining Trustworthy AI Agents
“In customer support, it’s critical that AI agents consistently reflect the company’s brand values, adhere to policies, and comply with regulations,” Upadhyay emphasizes. “We rigorously evaluate every AI model across all clients, both before and after deployment, focusing on five key performance areas.”
These five pillars-automation rate, execution accuracy, precision, response speed, and safety-constitute Zendesk’s comprehensive benchmarking framework. Each AI model is assessed on its ability to resolve issues correctly, follow instructions, respond promptly, and operate within strict ethical and operational boundaries. The objective is not merely to accelerate AI responses but to ensure they are dependable, responsible, and aligned with exceptional customer service standards.
Complementing this process is Zendesk’s quality assurance (QA) agent, an automated system that continuously monitors every AI-customer interaction. If a conversation deviates in tone or accuracy, the QA agent immediately flags the issue and alerts a human agent to intervene. This safeguard maintains a high-quality customer experience even when AI handles the initial support contact.
Advancing Support with GPT-5 Integration
The evolution from basic chatbots to autonomous AI agents capable of executing complex tasks marks a significant leap in customer service technology. For instance, an AI agent can now manage a product return by verifying eligibility, processing the return, and issuing refunds without human involvement. The launch of ChatGPT-5 presented Zendesk with an opportunity to embed these advanced capabilities into its Resolution Platform.
“GPT-5 represents a substantial upgrade in AI proficiency, transitioning from simple question answering to sophisticated reasoning and action-taking,” Upadhyay states. “It excels at independently solving problems, better understands customer intent-enhancing the feeling of being truly heard-and achieves over 95% accuracy in task execution.”
These improvements benefit Zendesk’s AI agents, Copilot, and App Builder tools. GPT-5 reduces workflow failures by 30% by maintaining context through complex scenarios and decreases fallback escalations by more than 20%, delivering more thorough and precise responses. This results in quicker resolutions, fewer transfers to human agents, and AI that operates more like an experienced support specialist than a scripted assistant.
Moreover, GPT-5 excels at interpreting ambiguous customer inputs, enabling it to clarify unclear requests and improve routing efficiency. This capability boosts automated workflows in over 65% of interactions. It also supports enhanced accuracy across five languages and generates concise, context-aware responses that align with brand tone guidelines.
In the App Builder environment, GPT-5 accelerates development workflows by delivering 25% to 30% faster performance and enabling more prompt iterations per minute.
Bridging the Gap in Support Analytics
Historically, customer support analytics have concentrated on structured data-metrics like ticket open times, agent assignments, resolution durations, and closure timestamps. However, the richest insights often reside in unstructured data: the actual conversations occurring via email, chat, voice calls, and messaging platforms such as WhatsApp.
“Many organizations underestimate the wealth of intelligence embedded in their support interactions,” Upadhyay remarks. “Our goal is to harness analytics to unlock insights that can enhance not just support teams but the entire business.”
To achieve this, Zendesk acquired HyperArc, a pioneer in AI-native analytics known for its proprietary HyperGraph engine and generative AI-driven insights. This acquisition revitalized Zendesk’s Explore analytics platform, transforming it into a sophisticated tool that integrates structured and unstructured data, supports conversational querying, and leverages persistent memory to contextualize new inquiries based on historical interactions.
“Support tickets reveal what’s broken in your business today-millions of data points accumulated over time,” Upadhyay explains. “Our enhanced AI agent can analyze this vast information and provide clear, actionable recommendations. This capability elevates support and drives improvements across the entire organization.”
With this newfound visibility, companies can identify recurring issues, uncover underlying patterns, and receive strategic guidance on resolution approaches. The system can even forecast potential problems before they arise. For example, during high-demand periods like Black Friday, it can analyze past data to detect frequent bottlenecks, predict emerging challenges, and suggest preventative actions-shifting customer support from reactive troubleshooting to proactive planning.
“This predictive power is where HyperArc truly excels,” Upadhyay concludes. “It not only helps you understand historical trends but also empowers you to prepare more effectively for the future.”
By embedding HyperArc’s AI-driven intelligence, Zendesk is pioneering a continuous learning approach in customer service, where every interaction builds trust and enhances performance, paving the way for AI systems that anticipate customer needs and business challenges.
