stack.tools

Zendesk

Customer and employee service software — ticketing, help center, and AI service agents.

Zendesk is unusually explicit about how it picks models: an internal testing system scores each new frontier model on 15 metrics using agents written the way customers write them, and it reruns the bake-off whenever a model ships. That process picked Claude Sonnet 4.6 for the custom agents Zendesk customers build for business-specific service work, argued on cost per task rather than token price. Serving runs through Amazon Bedrock in the AWS regions Zendesk already operates, keeping model processing inside its existing governance and data-residency story and sparing customers an extra cloud subprocessor. The custom agent builder itself was built by a five-person team working with Claude Code and the Claude Agent SDK, and the roadmap adds a router that sends simple requests to one model call and harder ones through a plan-then-act loop whose agents save what they learn as reusable skills.

zendesk.com
Facts
2
Receipts
3
Layers
2

Sources last verified Oct 12, 2026

Claude by Anthropic

Powers the custom service agents its customers build and run.

High confidence

Receipts · 2

Show quotes (2)
  • “The agents run on Claude, and they ran 1 million times in their first seven weeks of early access.” — claude.com, Oct 2026
  • “Zendesk chose Claude Sonnet 4.6.” — claude.com, Oct 2026

Source last verified Oct 12, 2026

Claude Code by Anthropic

Coding assistant its engineers used to build the custom agent builder.

High confidence

Receipts · 1

Show quotes (1)
  • “Built its custom agent builder with a five-person team, working with Claude Code and the Claude Agent SDK” — claude.com, Oct 2026

Source last verified Oct 12, 2026