At its AI summit on Wednesday, Zendesk introduced a suite of products powered by large language models, aiming to reduce the company’s dependence on human support staff.
The highlight of these updates is a fully autonomous support agent, which Zendesk claims can resolve 80% of customer inquiries without any human help. This tool will be complemented by a co-pilot agent to assist staff with the remaining 20% of cases, as well as additional agents for administrative tasks, voice interactions, and analytics.
Shashi Upadhyay, Zendesk’s president of Product, Engineering, and AI, explained that these new agents reflect a major transformation in the support sector, as artificial intelligence takes over many tasks traditionally handled by people.
“We’re moving from software designed for people to use, to systems where AI handles the majority of the workload,” Upadhyay said in an interview with TechCrunch.
Independent evaluations indicate that today’s AI models are up to the challenge. TAU-bench, a benchmark focused on assessing a model’s ability to use tools, features a scenario where models manage product returns—a task similar to many support roles. The top performer, Claude Sonnet 4.5, currently resolves 85% of these test cases.
Following a turbulent shareholder dispute in 2022, Zendesk has acquired several AI companies to support this transition. The analytics agent released today is based on technology from Hyperarc, which Zendesk acquired in July. Other recent purchases include Klaus, a QA and agentic service provider (acquired in February 2024), and the automation platform Ultimate (acquired in March).
Zendesk has already given some customers early access to the new system, and Upadhyay reports that the feedback has been very positive.
“Among users who have tried it, customer satisfaction scores have increased by five to ten points,” he told TechCrunch.
While large language models have been used in customer support before, few deployments match Zendesk’s scale. Companies like Airbnb and Regal Theaters have tested their own chatbot solutions, often working directly with foundational model developers. However, these systems usually focus on retrieving information rather than handling more complex troubleshooting or making autonomous decisions.
Should this new AI-driven approach to customer support succeed, the economic impact could be substantial. Zendesk’s Resolution Platform already serves nearly 20,000 clients and processes 4.6 billion support tickets annually. In the U.S. alone, there are 2.4 million customer service representatives, with even larger numbers employed globally.