How do you set up an AI agent in Zendesk? Six steps (2026)

Setting up an AI agent in Zendesk with Macha comes down to six steps: connect Zendesk (OAuth or an API token), pick a model (GPT-5.4 is the platform default), write plain-English instructions, add your Help Center as knowledge, attach a ticket.created trigger, and test on real tickets before the trigger goes live. None of it needs code, and a first agent is usually running in under 30 minutes.

Why put an AI agent in Zendesk at all?

Look at your last 100 tickets. A large share will be some variation of the same handful of questions: where's my order, can I get a refund, how do I reset my password, do you ship to my country. Count them yourself; the share varies a lot by business, and your own number is the one that matters. These tickets are repetitive, they drain your team, and the customer just wants an answer.

An agent that handles even half of that volume frees your team for the tickets where judgment and context matter. That's the whole pitch.

If you are still choosing a tool rather than setting one up, read how the alternatives price the same work first: our guides to Ultimate, now sold as Zendesk's own AI agents, and to Intercom Fin cover the two most common comparisons.

How do you connect Zendesk to Macha?

Click Connect, sign in to Zendesk and authorize. OAuth is the default path, and tokens refresh on their own, so you don't manage expiry.

If you'd rather use a sandbox or an API token, that path is still there. You'll need your subdomain and an admin-generated token. Either method takes under a minute, and the connector gives the agent 17 Zendesk tools covering tickets, custom fields, attachments, messaging and Help Center content.

One thing teams miss: you can connect multiple Zendesk accounts. If you run support for two brands on separate subdomains, name each connector instance (for example "Zendesk: Brand A" and "Zendesk: Brand B"). Tools are labeled by instance so the agent knows which account it's acting on.

Which AI model should the agent use?

Macha gives you a model picker per agent. The main options in the docs:

• GPT-5.4: the platform default today. Strong instruction-following for nuanced customer-facing replies, conditional policy logic, cancellations and billing disputes. Start here.
• GPT-5.6 Terra: the strongest option in the picker, for the agents where a wrong answer hurts most.
• GPT-5: the previous-generation all-rounder, still fine for agents with long, multi-step instructions.
• GPT-5.4 Mini: fast, for high-volume, lower-stakes work like triage, classification, summaries and field updates.
• Anthropic Claude: worth trying when the OpenAI models aren't giving you the writing quality or instruction-following you need on a particular agent.

Practical advice: start with the default, GPT-5.4, for your first agent. After a week, read the conversation logs. If the model is over-spec'd (most tickets resolved trivially), try GPT-5.4 Mini for speed. If it's under-spec'd (wrong answers, missed nuance), move up to GPT-5.6 Terra. Your price doesn't change either way, because Macha bills per ticket, not per model. The post on choosing the right AI model has the full decision framework.

How should you write the agent's instructions?

This is the decision that matters most in the whole setup. The instructions define what the agent does, what it doesn't do, and how it makes judgment calls.

Write them like onboarding notes for a new hire:

You're a customer support agent for [Brand]. When a customer asks about their order, look up the order in Zendesk and explain the current status. If the customer asks for a refund under $50, process it without escalation. If over $50, escalate to the team by adding a private note tagged "refund-review".

Always search the Help Center first before answering. If no article matches, look at the customer's previous tickets for context.

Tone: friendly but direct. Use the customer's name. Sign off with "Macha Support" but don't use exclamation marks.

Don't write code or pseudo-code. Write plain English rules and keep them under 500 words. You can iterate later, and if you'd rather start from a description, "Build with AI" drafts the name, instructions, tools and triggers for you, and "Edit with Sidekick" refines an existing agent through conversation.

How do you give the agent your Help Center?

This is where most setups go wrong: skipping knowledge. An agent without a knowledge source guesses. An agent with one cites your existing articles.

Click "Add knowledge source", pick Zendesk Help Center, and you're done. Macha indexes every published article and picks up new ones as they're published. If you unpublish an article, it's deactivated rather than deleted.

Search is hybrid (vector similarity plus keyword), so the agent finds the right article even when the customer's wording is nothing like the title. There's no retraining step when you edit docs. See the deep-dive on how Help Center auto-sync works for what happens underneath.

Which trigger should start the agent?

The trigger decides when the agent runs. The most common choice is ticket.created: every new ticket fires the agent, which reads the ticket, searches knowledge and replies or escalates within seconds. For Zendesk, Macha creates the webhook in your account automatically; you add a Zendesk trigger rule that points to it.

Other Zendesk events you can use:

• ticket.comment_added: the agent reacts when a customer replies, which keeps conversations moving
• ticket.status_changed: fires on transitions like open to pending, useful for follow-ups
• ticket.priority_changed, ticket.assigned and ticket.closed: for routing and wrap-up agents
• Custom webhook: wire any external system to start the agent

Attaching a trigger moves the agent from chat mode to autonomous mode, and in autonomous mode write actions run without the confirmation gate that chat has. The autonomous triggers guide covers setup in depth, including how to watch the first 24 hours before you trust the agent to run unsupervised.

How do you test the agent before it goes live?

Don't skip this. Macha gives you three ways to test, and they answer different questions:

• Try it: an interactive sandbox on the agent's page. Paste in a customer message and watch which tools the agent calls and what it would reply.
• Simulations: replay a batch of past Zendesk tickets with write tools stubbed, so nothing touches a live ticket. An AI judge gives a verdict on each conversation.
• Test run: Sidekick writes a realistic scenario, creates a real ticket in the connected workspace, runs the agent on it and reviews the result.

Run 5 to 10 tickets covering the range of cases you expect. Tune the instructions, re-test, repeat. That loop is the difference between an agent that works on day one and one that takes weeks to debug.

What should you check in the first 24 hours?

We don't quote a resolution rate here, because it depends on your ticket mix and knowledge coverage. What to look for on day one:

• Which categories the agent closes on its own, usually the repetitive ones (order status, refund eligibility, basic how-tos)
• First-response time on the tickets it handles, which should drop to seconds
• What it escalates, and whether those are the edge cases and upset customers you want a person on
• CSAT on AI-handled tickets against your human baseline

If the agent replies when it shouldn't, tighten the instructions. If it escalates too much, loosen them. The first week is iteration; by week two you should have an agent that earns its keep.

How we researched this: This walkthrough was written from Macha's own Zendesk connector and agent builder, and checked against Macha's docs (agents, triggers, connectors, simulations) on 24 September 2026. The model list and trigger events are the ones in the docs on that date; check the dashboard if something looks different.

How much does Macha cost for a Zendesk AI agent?

Plans start at $299/month for 750 tickets (about $0.40 a ticket): one plan, every feature, priced only on ticket volume, with unlimited agents and seats and setup by the Macha team. See full pricing, or install Macha for Zendesk from the marketplace and start your first agent today.