I started my career on an IT helpdesk. Password resets, access requests, printer exorcisms, the occasional "my laptop smells funny." I later built up that helpdesk function, and eventually ran the global IT organization it belonged to. So when I look at what AI agents are starting to do in IT Service Management, I don't see an abstract trend. I see my old job (the first five years of it) becoming something software does. That's not a lament. Most of what filled my queue back then never needed a human. It needed availability, consistency, and access to the right systems. Those happen to be exactly the things AI agents are good at. Gartner predicts that by 2028, one-third of enterprise software applications will include agentic AI, enabling 15% of day-to-day work decisions to be made autonomously. For anyone running IT operations, the question is no longer whether AI joins the service desk: it's what your operating model and your people do when it does.

From ticket queues to teammates

The current generation of ITSM agents is categorically different from the chatbots we bolted onto portals a few years ago. Those followed scripts. These understand context, plan a sequence of actions toward a goal, and learn from each interaction. In practice, that looks like:

• End-to-end ticket handling. Categorizing, routing, and often resolving requests without a human touch. At one financial services firm, an AI model classified over 500 tickets a day at more than 80% accuracy. The password reset (the classic hidden cost of every large IT organization) simply disappears as a human task.
• Copilots for the humans. Assistants that draft responses, summarize ticket history, and suggest the right knowledge article. The practical effect is that a first-year agent can operate closer to a seasoned one's level, the kind of leveling-up that used to take years of sitting next to the right colleague.
• AIOps and self-healing. Agents that watch logs and metrics around the clock, flag the database latency spike at 2 AM, and, where a known fix exists, apply it: restart the leaking process, roll back the bad deployment, fail over before users notice. ServiceNow now ships agents that troubleshoot network issues "like a Tier-3 engineer would."
• AI in change and problem management. Mining historical change data to generate implementation, test, and backout plans, or clustering incidents to hypothesize root causes. This is AI moving into ITIL territory that has always been reserved for experienced humans.

The direction of travel is what some in the industry call the "zero ticket" future: issues detected and resolved before a user ever writes one. Having spent years staring at queue dashboards, I can tell you that the queue was never the point. The point was that something was broken. If it can be fixed before anyone notices, nobody misses the ticket.

What actually changes in the operating model

ITIL doesn't become obsolete, but the tiered-support pyramid I grew up in does get rebuilt. The classic model put the cheapest, least experienced people at Tier 1, handling the highest volume. When agents absorb that volume, the humans who remain aren't operators anymore. They're orchestrators: managing exceptions, supervising AI-driven workflows, and improving the system that does the work. That shift is bigger than it sounds. A Tier-1 agent's job stops being "resolve the ticket" and becomes "train and tune the thing that resolves the tickets." An incident manager spends less time logging and coordinating, more time on the patterns the AI surfaces. And new roles appear that didn't exist in any ITIL book: someone has to own the knowledge the agents draw from, and someone has to watch the watchers. Gartner expects that by 2028, 40% of CIOs will demand "guardian agents" to track and contain the results of autonomous AI actions. The trust boundary matters more than the org chart, and the data here is sobering in a healthy way: over half of IT professionals say they don't trust AI to make decisions without human oversight. That instinct is correct, and the operating model should encode it: clear rules for where agents act freely (password resets, routine diagnostics), where they need sign-off (production changes), and full audit logs of everything they do, reviewed the way we review human-caused incidents.

The skills that matter now

If I were starting on a helpdesk today, I'd invest in a different toolkit than the one I built in 2010:

• AI interaction design. Crafting the prompts, dialogue flows, and escalation logic that make agents perform. This is the new "knowing the ticket system inside out."
• Data fluency. Agents generate mountains of operational data. The professionals who can interpret it (validate the AI's findings, spot the trend behind the anomalies) become the ones who improve the system rather than just operate it.
• AI governance and auditing. Knowing how to ask "is our classification model making correct and fair calls, and how do we catch its mistakes?" In a 2025 survey, lack of AI expertise and governance concerns were the top barriers to adoption in ITSM: the skills gap is the bottleneck, not the technology.
• The human layer. With machines handling the mechanical work, empathy, communication, and judgment stop being soft skills and become the differentiating ones. Users will forgive a bot for being a bot; they won't forgive an IT organization that has no humans left when it matters.

The organizations getting this right treat it as a workforce transition, not a tooling upgrade. The person who spent five years resolving tickets holds exactly the domain knowledge the agents need to be trained on; throwing those people away because "AI does Tier 1 now" is burning the training data with the trainer.

The caveats I'd put in the business case

Two things temper my enthusiasm, both from experience rather than theory. First, over-automation is a real failure mode. A badly trained bot that confidently gives wrong answers doesn't reduce workload; it adds a new incident category and burns user trust that took years to build. Users have rational expectations (speed, availability) and emotional ones (being heard when something is genuinely wrong). AI is excellent at the first and, for now, mostly theater at the second. Design the handoffs accordingly, and be transparent: people should know when they're talking to a bot and have an obvious path to a human. Second, accountability doesn't automate. When an agent auto-closes the wrong ticket or a remediation script takes down a service, "the AI did it" is not an acceptable line in the post-incident review. Someone owns the outcome. Log every agent action, review AI-driven decisions the way you'd review a new hire's work, and keep sensitive categories (HR tickets, legal matters) out of the agent's reach until you've earned confidence deliberately. The service desk was where I learned that good IT is mostly about understanding what people actually need. That lesson doesn't change when half the team is software. The tickets were never the job; they were the interface to the job. AI teammates are about to handle the interface better than we ever did, which leaves the actual job, understanding and improving how the organization works, squarely with us.