
Zendesk's August 2026 release notes contain a sentence worth reading twice: "Copilot intelligent triage is now included in Suite and Support Professional plans and above, enabling automatic AI classification of incoming tickets by topic, sentiment, language, and entities at no extra cost."
A capability that used to sit behind an AI add-on is now bundled into the plans most Zendesk customers already pay for. That is genuinely good news if you're on one of those plans, and credit where due: automatic classification by topic, sentiment, language and entities is useful, and the release also mentions workflow recommendations built on top of it.
But the more interesting question for anyone evaluating helpdesk AI is not what became free. It's why this particular thing became free — and what stayed paid.
Zendesk's Copilot add-on — described on their pricing page as "AI assistance that empowers your entire team" — is still a separate line item: $50 per agent per month, billed annually, on Professional plans and above (checked at zendesk.com/pricing, August 2026; monthly billing is a different, higher rate). It sits alongside a Workforce Engagement bundle at $50 per agent per month and an AI-powered Contact Center at $83 per agent per month, both also billed annually.
Notice the unit in every one of those prices: per agent. Per human. The AI charge is counted against the size of your team.
There's a familiar structure here. When a vendor gives away one layer of an AI stack, it's usually the layer that makes the paid layer more valuable. Free triage means every ticket in your queue arrives pre-classified — which makes the case for the paid assistance on top of it stronger, and makes leaving the ecosystem harder. The razor is free; the blades are the subscription.
Nothing about that is dishonest. It's just worth seeing clearly, because the moment matters: Zendesk is currently walking customers through a series of AI migrations — technical development on its legacy AI agents ends August 31, 2026, with a full shut-off on December 10, 2026. Expanding free access to the new triage while retiring the old generation is exactly how you move an installed base. If you're being moved, it's a good time to ask where you're being moved to, and what the destination costs at your team's size.
1. What is the AI charge counted against? Per human agent, per resolution, per session, or per AI? A per-agent price means hiring two more support people raises your AI bill whether or not those two ever touch the AI. A per-resolution price (Intercom's Fin is $0.99 per resolution) means a busy month costs more than a quiet one. Each model suits somebody — but you should know which one you're in before your team grows.
2. Does the free part act, or only inform? Classification tells you a ticket is urgent, angry, and about billing. Somebody — or something — still has to answer it. When you model costs, price the part that produces the reply your customer actually receives.
3. What happens to the price when you grow? Run the arithmetic at your headcount next year, not this year. Ten agents on a $50-per-agent add-on is $6,000 a year for the AI alone, before the seats themselves. The same math at a per-AI price doesn't move when you hire.
DeskLeap's answer to question one is different by design: the platform — shared inbox, ticketing, live chat, knowledge base, customer portal — is 100% free, with unlimited human agents. The only paid unit is an AI Agent at $29/month each, with 500 AI replies included per agent monthly. It answers chats and tickets from your knowledge base, and it's priced per AI, not per teammate. Growing your team doesn't change your bill; adding another AI does, by exactly $29.
That model has trade-offs too — if you want ten AI agents, you pay for ten. We think counting the robots instead of the humans is the honest direction for this market, and pricing moves like free triage — where the informative layer goes free while the per-human meter stays — make the contrast easier to see, not harder.
Either way: when the next "now free!" announcement lands in your inbox, find the unit on the part that still costs money. That's where the pricing model actually lives.