Zendesk claims its new AI agent can resolve 80% of support issues

Zendesk announced a series of LLM-based products Wednesday at its AI summit aimed at reshaping the company’s reliance on human technicians.

The center of the new features is an autonomous support agent that Zendesk says will resolve 80% of support issues without human intervention. This system will be complemented by a co-pilot agent, which will assist human technicians with the remaining 20% ​​of problems, as well as an administrator layer agent, a voice agent and an analytical agent.

According to Shashi Upadhyay, Zendesk’s president of product, engineering and AI, the new agents are part of a broader shift in the support industry, as AI replaces much of the work previously done by humans.

“The world is going to move from software designed for human users to a system where AI does most of the work,” Upadhyay told TechCrunch.

Independent benchmarks suggest that contemporary AI models are capable of taking on this task. TAU-bench, designed to measure a model’s ability to invoke tools, includes a scenario in which models must process a returned product – a close analogue of many support tasks. The current leader, Claude Sonnet 4.5, fixes 85% of the problems in the test.

After a chaotic investor fight in 2022, Zendesk made a series of AI acquisitions that laid the foundation for today’s change. The analytics agent launched today builds directly on the company’s acquisition of Hyperarc, completed in July. Previous AI acquisitions include quality assurance and agent services system Klaus (acquired in February 2024) and automation platform Ultimate (acquired the following March).

Zendesk has tested the new system with its existing customers and Upadhyay says the results are promising.

“For customers who use it, consumer satisfaction increased by five to 10 points,” he told TechCrunch.

Large language models have often been deployed for customer support, although rarely at the scale of Zendesk. Companies, like Airbnb and Regal Theaters, have already experimented with in-house chatbot support, often contracting directly with foundation model labs. But these systems are generally concerned with retrieving information rather than more complex troubleshooting or taking self-directed actions.

If the new initiative for AI-based support succeeds, the economic implications would be significant. Zendesk’s resolution platform already supports nearly 20,000 customers, resolving 4.6 billion tickets each year. Beyond Zendesk, the United States employs 2.4 million customer service representatives, with much larger workforces in other countries.