You’re reading an issue of “The AI Economy,” my newsletter exploring the forces shaping the AI era—tracking how AI is rewriting business, work, technology, and culture. Subscribe to get expert insights and curated updates delivered straight to your inbox.
Zendesk spent years building software to manage customer service tickets. Now, its attention has turned to building an AI workforce that resolves them. It’s a “fundamental reset” of its platform, one where deflection-based bots are replaced by specialized AI agents that operate across all channels.
At this year’s Relate conference, Zendesk is unveiling autonomous AI agents for Employee Service; an expanded agent portfolio across all customer channels; a new copilot suite for human teams; a no-code Agent Builder; continuous quality measurement; deeper knowledge infrastructure; and support for the Model Context Protocol. To top it all off, the company is introducing an enhanced version of its Outcome-Based Pricing model, one that supports double verification.
“We are entering the age of the autonomous service workforce,” Tom Eggemeier, Zendesk’s chief executive, said in a statement. “We believe the day is coming when every great business will be powered by specialized AI agents that don’t just work for you, but work with you.” He added that the agents will behave like real “team members,” with the same level of accountability—it’s a pitch that sounds familiar from nearly any enterprise AI vendor right now, such as Asana, Salesforce, and Microsoft.
Despite Eggemeier’s pronouncement, today’s announcements lean heavily on infrastructure upgrades and human-assist tooling. But that reading misses the point. Zendesk laid the foundation last year with its Resolution Platform—the unified layer that brings together data, intelligence, knowledge, workflows, and governance—and what’s being added now are the capabilities that make autonomy operational. A few targeted additions carry more strategic weight than the full product list implies.
The autonomous service workforce only works if agents can operate without handoffs to humans. This happens when there’s a loss of context. Every time a customer moves from chat to email to a phone call and has to re-explain their problem, that’s a failure point. To address this, Zendesk reveals that its AI agents now operate across messaging, email, and voice with shared context across interactions. Agent amnesia is an industry-wide problem, one that companies like Twilio have also moved to address. For Zendesk, this is made possible through its acquisition of Forethought in March.
Cross-channel continuity is only part of the equation. Zendesk is also expanding multilingual support for its Voice AI Agents. These bots can now converse in more than 60 languages and switch mid-conversation without losing context, a capability that could go a long way toward making support calls less dreadful. The expanded language support benefits customers directly, but it also keeps Zendesk competitive against a growing field of voice-native challengers like Sierra, Dialpad, Fin (formerly Intercom), Google, Twilio, and Salesforce—all moving aggressively to reshape how businesses talk to their customers.
Zendesk’s ambitions for an autonomous service workforce don’t stop at the customer. It also extends inward. The CX platform is furthering its march into territory traditionally dominated by vendors like ServiceNow and Moveworks. It’s introducing autonomous AI agents as part of its Employee Service offering, a platform introduced at last year’s Relate conference. Powered by its Unleash acquisition, these agents operate inside Slack and Microsoft Teams, performing functions similar to those of Atlassian’s Rovo and Glean, searching across enterprise systems to provide workers with answers to their queries based on the systems they’re authorized to access.
And for organizations wanting to build their own agents, Zendesk has introduced Agent Builder, a no-code tool for building, testing, deploying, and optimizing custom agents. The company said it’s designed to help “automate more complex front, middle, and back office service work while maintaining governance and oversight from a single control plane.” That said, it’s not the only agent builder that Zendesk has—it unveiled a similar offering last year as part of its Resolution Platform. The status of that one remains uncertain. In addition, for developers wanting greater flexibility, Zendesk is unveiling an MCP client and server.
“Our mission is to put the power to create this workforce into the hands of every enterprise, on one elegant platform,” Eggemeier said. “Whether those agents are crafted by Zendesk, by our partners, or by your own teams, they will all speak with one voice. We are not just giving you a tool; we are giving you the future of work.”
Here’s a look at other announcements from Zendesk Relate:
The Context Graph
Zendesk is debuting its Context Graph, an operational memory layer that captures past analyses, performance context, and agentic reasoning to improve future recommendations.
And speaking of graphs, Zendesk is expanding its Knowledge Graph by adding connectors for SharePoint, Google Drive, Notion, Guru, Contentful, and Document360.
Enhanced Copilot Experiences
Not everything Zendesk announced Tuesday is about replacing human effort. The company also unveiled a suite of Copilots, tools designed to work alongside human agents rather than supplant them. The Agent Copilot connects to internal and external sources to generate procedures and take action on at least 30 percent of tickets from day one. The Admin Copilot helps administrators identify operational issues, recommend fixes, and apply workflow changes in real-time.
For knowledge teams, the Knowledge Copilot surfaces gaps, outdated content, and inconsistencies based on actual customer conversations. And the Analyst Copilot helps teams spot trends and determine root causes through an agent-powered analytics experience.
Measuring What Matters
Rounding out the human-assist portfolio is Quality Score, a new continuous measurement feature that analyzes 100 percent of human and AI interactions to give teams an objective view of service quality and surface opportunities for improvement in real time. It’s available exclusively on Suite Professional plans and above.
Doubling Down on Outcome-Based Pricing
Introduced in August 2024, Zendesk’s outcome-based pricing charges customers only for resolutions that are verifiably resolved. Tuesday’s update reinforces that model with a double-verification process. Every resolution is first confirmed by the AI agent, then independently by a separate evaluation model. It’s an approach that echoes what Microsoft is doing with its Microsoft 365 Copilot, and a signal that verification standards could become a competitive battleground in enterprise AI pricing.
All of this is Zendesk making good on a promise. In December, Eggemeier told The AI Economy it was time for Zendesk to be the disruptor, not the disrupted. The company generated $200 million in AI annual recurring revenue last year and is targeting up to $500 million this year. The autonomous service workforce is how it gets there—lower costs and better outcomes for customers translates directly into higher retention and expanded spending for Zendesk.
Not everything announced at Relate is available today. Agent Builder, the Context Graph, the MCP Client, and the Analyst and Knowledge Copilots are in early access now. The updated Voice AI Agents reach general availability later this quarter, while the Employee Service agents and MCP Server open to early access this summer. The Admin Copilot and expanded Knowledge Graph connectors are generally available today. Quality Score has no confirmed timeline beyond a vague “coming soon to early access.”
