Search for a Zendesk AI agent and you mostly get Zendesk, which is fair enough given it is their product, and not much help if you are trying to work out what to actually turn on. There are three genuinely different routes. The choice between them is not really about features, it is about one question: does the work you want automated finish inside the ticket, or does it finish somewhere else?
The three ways to run an AI agent on Zendesk
Every option in this category collapses into one of three shapes, and they differ mainly in how far the agent can reach beyond the helpdesk. Pick by scope first, then compare features inside the shape you picked.
| Route | Best when | Reaches beyond Zendesk? | How it is billed |
|---|---|---|---|
| Zendesk's own AI agents | The question is answerable from your help centre and ticket history | No, except through Zendesk's own action builder | Per agent seat, plus per automated resolution |
| Marketplace app | You want one narrow behaviour added to Zendesk | Rarely, and usually only to one other system | Usually per seat or flat, on top of Zendesk |
| External agent platform | The answer lives in billing, a spreadsheet, a warehouse or email | Yes, it holds credentials for each tool | By the work, independent of helpdesk seats |
What Zendesk's own AI agents do, in Zendesk's words
Worth reading the vendor's own description before anyone else's summary of it. Zendesk's pricing page says its AI agents are "powered by agentic AI" and that they "reason, adapt and take action, then learn and self-improve to achieve full resolution", while following your business processes.
The page also names the pieces you get at each tier. Suite Team lists AI agents, an AI knowledge base and an AI action builder, which Zendesk describes as a way to "design, integrate and automate multi-step action flows across Zendesk and third-party systems". Suite Professional adds an admin copilot that "proactively surfaces personalised insights and recommends workflow and set-up improvements", plus an AI app builder, AI writing tools and AI quick reports. Zendesk separately markets Forethought AI agents by Zendesk, which it describes as self-improving agents that work inside Zendesk but also on any platform.
The honest summary: the built-in option is strong at answering from content you already have in Zendesk, and its reach into other systems depends on how much you are willing to build in the action builder.
How Zendesk actually bills an AI agent
This is the part that surprises people, so it is worth being precise. Zendesk sells seats and AI usage as two separate meters, and the second one is priced on outcomes rather than on volume of attempts.
Zendesk states that its AI agents are "included in every Suite and Support plan, with pricing based on the successful outcomes they deliver", and defines the unit plainly: automated resolutions are "the unit of measurement used for calculating and billing your account for AI agent usage". It adds that paying per automated resolution "means that you pay only for customer requests that were successfully resolved by the AI agent, without any escalation to a human agent".
That is a genuinely fair-sounding model, and it has one consequence worth planning for: your AI bill rises as the agent gets better. A deflection is the billable event. An agent that resolves nothing is free and useless; an agent that resolves half your volume is working and expensive. Budget for the success case, not the pilot.
Here is what the seat side costs before any AI usage, taken from the plans listed on Zendesk's pricing page on 10 September 2026.
| Plan | Price, billed yearly | AI included in the plan listing |
|---|---|---|
| Support Team | US$19 per agent per month | Ticketing, routing, automations and triggers |
| Suite Team | US$55 per agent per month | AI agents, AI knowledge base, AI action builder |
| Suite Professional | US$115 per agent per month | Admin copilot, AI app builder, AI writing tools, AI quick reports |
| Suite Enterprise + Copilot | Quoted | Positioned for advanced security, governance and the most proactive AI assistant |
Doing the arithmetic on those published numbers: a five-person support desk on Suite Team is US$275 a month in seats before a single automated resolution is billed, and the same desk on Suite Professional is US$575. What we cannot tell you is the third number, because Zendesk does not print it. We read the pricing page on 10 September 2026 and it defines the automated-resolution unit without publishing a rate next to it, and it points to a savings calculator rather than a price. Several third-party posts circulate a per-resolution figure; we are not repeating it, because it is not on the vendor page and we could not verify it. Get the rate quoted before you model the rollout, and model it at your realistic deflection rate rather than your pilot one.
Which Zendesk work is actually worth automating
The pattern that holds across helpdesks: automate the work where the input is already structured and the output is a draft. Those tasks pay back quickly and fail visibly, which is the combination you want.
- Ticket triage and routing. Reading an inbound ticket, tagging it, setting priority and sending it to the right queue. High volume, low judgement, easy to check.
- Macro suggestion. Surfacing the right saved reply to the human who will actually send it. The human stays in the loop and the time saving is immediate.
- First-reply drafting. Producing a draft from your help centre content, held for review on anything sensitive.
- Tagging and deduplication. Merging the four tickets that are the same outage, and keeping tags consistent enough that your reporting means something.
- Quiet-ticket chasing. Finding tickets waiting on a customer reply and nudging them, which is pure housekeeping nobody misses doing.
The tasks to leave alone are the ones that move money or entitlement: refunds, plan changes, cancellations, credits and anything touching an account's access. Not because an agent cannot execute them, but because those are exactly the actions a customer notices when they are wrong, and reversing them costs more than the automation saved.
Set the permission model before you connect anything
Decide what the agent may do before you decide which agent to use. The useful line is between proposing and applying, and it is worth drawing it explicitly rather than discovering it after an incident.
- Let it write drafts, comments and internal notes. Reversible, reviewable, and where most of the time saving actually is.
- Let it tag, route and set priority. Cheap to correct if it gets one wrong.
- Hold back send-to-customer on anything sensitive. A draft a human approves captures most of the benefit at none of the risk.
- Never grant refunds, cancellations or entitlement changes unattended. Put a confirmation step in front of every one.
- Decide what happens on low confidence. The correct default is escalate to a human, not guess. This also matters commercially, since an escalation is not a billable automated resolution.
Task-by-task guides
We have written up the two Zendesk tasks that come up most often, each with the setup and the failure modes.
- AI agent for Zendesk ticket triage, on reading inbound tickets and routing them without mangling your queues.
- AI agent for Zendesk macro suggestions, on surfacing the right saved reply to the person about to send it.
- Best AI agents for customer support, if you are still choosing a platform rather than a task.
How to choose without a long evaluation
You can settle this in an afternoon with three questions, in this order.
- Where does the answer live? If it is in your help centre and ticket history, the built-in option is the shortest path. If it is in billing, a spreadsheet or a warehouse system, you need something that can reach those, and no amount of helpdesk AI will get you there.
- What is your realistic deflection rate? Because that is what you are billed for. Model the cost at the rate you expect after six months, not the rate in week one.
- How many seats versus how many people actually want this? Per-seat AI tiers price everyone on the desk, which is fine at five agents and much less fine at fifty when three people wanted the feature.
Vendor AI in a helpdesk is normally tied to plan tier or sold per seat, so cost scales with headcount rather than with the automation you actually use, and the outcome meter then scales with success. An external platform priced by the work is usually cheaper for a team where a handful of people would use it. Gravity has a free tier at $0 a month for one agent, and paid plans from $20 a month with a lot of usage included; buy more usage if you run out. If you are new to the category, what is an AI agent covers the basics, and AI agents for every profession maps the work by role.
Frequently asked questions
Is there an AI agent for Zendesk?
Yes, and there are three separate routes. Zendesk ships its own, sold as AI agents inside the Suite and Support plans, which its pricing page describes as agentic AI that can reason, adapt, take action and self-improve toward full resolution. The second route is a Marketplace app that adds agent behaviour to Zendesk itself. The third is an external agent platform that holds credentials for Zendesk alongside your other tools. Which fits depends on whether the work you want automated ends inside the ticket or somewhere else entirely.
How does Zendesk price its AI agents?
By outcome rather than by seat. Zendesk states that AI agents are included in every Suite and Support plan, "with pricing based on the successful outcomes they deliver", and it defines automated resolutions as the unit used for calculating and billing AI agent usage. Zendesk also states that paying per automated resolution means you pay only for requests the AI agent resolved without escalating to a human. Note that this sits on top of per-agent seat pricing, so an AI rollout adds a second meter rather than replacing the first.
What is an automated resolution in Zendesk?
It is Zendesk's billing unit for AI agent usage. Zendesk defines automated resolutions as the measurement used for calculating and billing your account for AI agent usage, and says that paying per automated resolution means you pay only for customer requests that were successfully resolved by the AI agent, without any escalation to a human agent. The practical consequence is that a deflected ticket is a billable event, so your AI bill scales with how well the agent works, not with how many people you employ.
Which Zendesk plans include AI agents?
Zendesk's pricing page lists AI agents from Suite Team at US$55 per agent per month billed yearly, alongside an AI knowledge base and an AI action builder. Suite Professional at US$115 per agent per month billed yearly adds admin copilot, an AI app builder, AI writing tools and AI quick reports. The entry Support Team plan is US$19 per agent per month billed yearly. Zendesk also states that AI agents are included in every Suite and Support plan, with the usage itself billed per automated resolution.
What Zendesk work should an AI agent handle first?
The work where the input is already structured and the output is a draft rather than a final action: triaging and routing incoming tickets, suggesting macros to the human who will send them, drafting first replies for review, tagging and deduplicating, and chasing tickets that have gone quiet. Keep refunds, cancellations, account changes and anything that touches money or entitlement behind a human confirmation step, because those are the actions a customer notices when they go wrong.
Do I need Zendesk's own AI to run an agent on Zendesk?
No. Zendesk's AI is the shortest path when the work starts and finishes inside the ticket, because there is nothing to integrate and the context is already there. If the work crosses into billing, a spreadsheet, a warehouse system or email, an external agent platform that holds credentials for all of those is usually the better fit. A Zendesk-only agent cannot see the system where the actual answer lives, and support questions very often resolve somewhere other than the helpdesk.