Zendesk AI: Everything You Need to Know
Zendesk AI agents are AI-powered chatbots that interact with your customers on messaging, email (including API and web form), and voice channels. They can have conversations with customers and perform actions in authorized systems autonomously, automating and resolving issues so human agents can focus on more complex work. AI agent functionality is available on all Zendesk Suite plans (Team, Growth, Professional, Enterprise, and Enterprise Plus) and Support plans (Team, Professional, and Enterprise). Usage of AI agent functionality is measured through automated resolutions: each account includes an allowance of automated resolutions based on its plan, and you can buy more if needed. This means you pay only for customer requests that were successfully resolved by an AI agent, without any escalation to a human agent.
What Zendesk AI agents are
Zendesk AI agents are customer-facing bots that reason, take action and resolve a request end to end without a human. They use agentic AI to solve complex user requests with limited direct human supervision, and are capable of autonomous decision-making, planning, and adaptive execution throughout a conversation. They interact in a natural, human-like way, engaging in small talk and working through vague issues by identifying the next steps and asking follow-up questions where needed.
One distinction matters across the whole portfolio: AI agents talk to the customer and can close a ticket with zero human involvement. Copilot (Agent, Admin, Knowledge, Analyst) always sits beside a person. A newer third category, Custom Agents, works behind the scenes on back-office tasks and can be called by an AI agent or by Copilot. Custom agents can now be invoked from AI agents through action flows without the request instantly timing out.
Zendesk is simplifying AI agent packaging and expanding access to its most advanced, agentic capabilities across all Suite and Support plans. This change was announced on March 30, 2026, with a phased rollout that starts on May 11, 2026 and ends on June 12, 2026. The rollout removes the distinction between "Essential" and "Advanced" AI agent plans, moving all customers to a single offering that includes agentic reasoning, multi-step procedures and external API integrations, plus a streamlined self-service onboarding flow and unified management across messaging, email, and voice (in EAP). This streamlined experience is powered by the Resolution Platform. While guided self-service is available for simpler use cases, expert guidance through AI Expert services remains critical for successful deployment at scale for more complex implementations. As part of this evolution, Zendesk plans to align pricing with the value customers receive and will share more details separately about how outcome-based pricing will evolve.
Rollout windows:
• New customers and trial accounts: May 11–15, 2026
• Existing customers who never created an AI agent, or who use essential or legacy AI agent functionality: May 25–June 12, 2026
• Existing customers using advanced AI agents: the current experience is unchanged; streamlined onboarding rolls out on a future timeline
The following pieces of functionality are now legacy:
• AI agents - Essential: legacy as of May 11, 2026. Only customers who had any drafted or published AI agents as of this date can still access it. It stops receiving bug fixes/technical upgrades on August 31, 2026 and is no longer available on December 10, 2026.
• Bot builder, answers, and intents: legacy as of February 2, 2025. Customers with any drafted or published AI agents as of that date retain access. Technical development stops (except critical bug fixes) on August 31, 2026, and these features are no longer available on December 10, 2026.
• Zero-training AI agents: legacy as of May 11, 2026. They can no longer be created, but existing ones can be maintained. Much of their configuration (such as use cases) carries over to the current agentic generation.
• Autoreplies with articles: legacy as of July 31, 2025. Technical development stops (except critical bug fixes and supporting breaking changes) on August 31, 2026, and this feature is no longer available on December 10, 2026.
Forethought AI agents by Zendesk
Forethought AI agents by Zendesk is an independent suite of omnichannel AI tools designed for customer support teams, including those that don't use the Zendesk platform. Its capabilities help organizations handle support requests more efficiently by automating common tasks, surfacing helpful insights, and improving the quality of both human and AI-assisted support.
How an AI agent is built
Zendesk describes every AI agent as four building blocks, orchestrated by a multi-agent architecture behind the scenes.
| Building block | What it does |
| Knowledge | The baseline the agent "just knows": help center plus any knowledge connector source (e.g., Confluence, Google Drive, SharePoint, Notion, Box). AI agents and generative search can also use PDF content. Underpins every automation. |
| Procedures | Step-by-step logic for a use case, written in plain language on top of the knowledge. Supports hybrid flows combining scripted paths with generative fallback. |
| Actions | Lets the agent do things, not just answer: API calls, custom actions, Action Flows / Action Builder no-code automations, and custom CRM integrations. |
| Analytics | Tracks resolutions per use case and channel, and shows the bot's reasoning for continuous improvement |
The agentic architecture splits the work across several internal agents: some clarify the customer's intent, some decide whether to answer from knowledge, ask a follow-up or trigger a procedure, and others carry out the actions. Zendesk's line is that it manages this complexity so admins only write plain-language instructions.
Inside the admin UI, each AI agent has three layers:
• Identity: name, tone of voice, pronoun formality (for example formal vs informal German), custom instructions and languages
• Knowledge sources: help center, knowledge connectors, web crawler; scoped with search rules and segments
• Use cases: a name and description (used to detect the intent) plus a generative procedure that says what to do
At a high level, building an AI agent follows this workflow: optimize your help center content for AI agents, configure the channels you want the agent to work on, create the AI agent, add use cases with generative procedures or dialogues, add actions, API integrations, and entities to increase automation, activate the agent on your chosen channels, and monitor its performance.
AI agent capabilities
| Feature | What it does |
| Generative procedures & hybrid flows | Write a use case in free text; the agent follows it, asks for missing info and calls actions. Supports hybrid flows combining generative fallbacks with scripted paths (e.g. order-status checks). One procedure covers every supported language. |
| Procedure diagram & builder view | Auto-generated flow showing how the AI interpreted your plain-language instructions (name, description, and procedure text); edit by chatting with it or modifying directly. |
| Intent disambiguation | Handles vague asks (e.g., "where's my money?") by checking what its knowledge and use cases cover and offering relevant paths (deposits, withdrawals, or missing funds). |
| Slot filling and multi-turn reasoning | Spots missing details (e.g., clarifying "lately" to mean the past month) and asks follow-up questions across multiple turns to calculate figures or complete operations. |
| Flexible data capture | Collect info via structured forms, carousels, or conversational free text depending on the channel; allows users to select specific items or specify choices in natural language. |
| Input validation | An agent can validate the length of numeric or alphanumeric inputs (such as order IDs or driver's license numbers) using a dedicated tool, improving reliability in verification and data-collection flows. |
| In-channel authentication | Verifies user identity on unauthenticated channels (e.g., WhatsApp, SMS) using CRM lookups, PINs, or OTPs before sensitive account updates. |
| Generative knowledge replies | Answers from help center and external knowledge sources (Google Drive, Confluence, SharePoint, Notion, Box), including PDF content, with sources cited and shown (now in a separate view on Web Widget and social). |
| Search rules and segments | Restrict which knowledge subset a customer gets based on conversation metadata or to deliver concise answers by customer type without creating additional use cases. |
| Reusable actions & Action Builder | Connect to external systems with no-code action flows and API integrations, so the agent can perform tasks and update data outside of Zendesk during a conversation. |
Agentic AI for advanced email AI agents
As of April 20, 2026, agentic AI for advanced email AI agents is generally available, and all newly created advanced email AI agents use agentic AI by default. Powered by fully generative AI, email AI agents can understand incoming emails, answer questions from your knowledge sources, and execute business procedures through system integrations without complex flow-building. Note that email AI agents can't use dialogues and must use generative procedures instead.
Key capabilities:
• Single and multiple question handling: answers one or more questions and aggregates all answers into one email to minimize back-and-forth.
• Procedure automation: executes generative procedures and collects all needed information from the customer upfront in a single email.
• Clarifying questions: asks for specifics when a customer's request is vague.
• Small talk recognition: recognizes small talk and responds appropriately.
• Escalation handling: escalates to a human agent when the customer requests it, when a procedure specifies it, or when it can't confidently resolve the request in one response.
• Messaging-aligned flows: automation flows, conversation status management, and business logic closely mirror Zendesk messaging.
• Automatic email cleanup: strips previous replies, signatures, footers, and legal notes so the agent focuses only on relevant new content.
• Automatic multi-language support: supports multiple languages out of the box, though you still need to add and activate the languages you want to support.
Employee Service AI Agents (EAP)
The Employee Service AI Agents EAP was announced on July 10, 2026, with rollout starting August 3, 2026. It delivers conversational AI-based support directly to your employees, resolving internal requests within their existing workflows. It turns on automatically for Employee Service Suite accounts, with exceptions for HIPAA or FedRAMP certified accounts, accounts opted out of generative AI, accounts that already have an active AI agent solution, and Employee Service Suite Team plans.
Key features:
• Enterprise search: connect internal knowledge with pre-built connectors to Google Drive, SharePoint, Confluence, Notion, Slack, Teams, Box, Seismic, and a Web Crawler.
• Secure knowledge: AI agents respect existing content permissions at the source, so employees only see what they're allowed to see.
• Employee channels: the AI agent can be integrated with Slack and Teams so employees get support where they already work.
• AI agent actions: soon after the EAP begins, an update will allow connecting the employee service AI agent to action flows to handle tasks across Zendesk and third-party services.
During the EAP, employee service AI agents are free to use. When the feature becomes generally available, resolutions using connected external sources might count toward your automated resolution credits and totals. This EAP is available on the Employee Service Suite only, but will expand to include eligible customers using the Customer Service Suite for internal service when the product becomes generally available.
Voice AI agent enhancements
Recent releases significantly expanded the voice configuration capabilities of Voice AI Agents:
• Choice between multilingual male and female voices.
• Voice overrides for specific locales and accents using higher-fidelity voice models (for example, one voice for English US, another for English UK, and a different voice for French or German customers).
• A new Hangup action that can be used within voice flows and procedures to cleanly conclude the conversation and end the call.
• An upgraded voice cancellation model rolled out across all conversations by default, improving audio quality, transcription quality, and conversational naturalness.
• Added pronoun-handling examples for previously unsupported locales, improving response quality and language consistency.
Developer and API capabilities
Developers can further customize and extend AI agents using the AI Agents APIs:
• Integrate bots with custom CRMs, business systems, and third-party APIs.
• Store and reuse session parameters and metadata tied to each user or session.
• Use webhooks to receive real-time event notifications and trigger backend actions.
• Escalate conversations to human agents with full context or implement custom escalation logic.
• Tune conversation flows with custom templates and message types.
• Access conversation data programmatically to power analytics, reporting, and compliance initiatives.
These connections use API-key authentication to securely connect to the AI agent platform.