Agentic AI will soon run your service desk. The real question is who controls it

In December, ServiceNow completed its US$2.85 billion acquisition of the AI company Moveworks - the largest deal in its history. For most IT leaders it registered as another headline in a noisy year. It deserves more attention than that. It marked the moment the service management industry crossed a line it cannot uncross. After three decades of building software that records work, we are now building software that does it. That shift is far more consequential than the chatbots and copilots that dominated the previous two years, and most organisations are about to respond to it in one of two wrong ways.

The first mistake is to dismiss it as hype. The second is to trust it uncritically. Both are understandable, and both are dangerous.

Consider what has actually changed. For thirty years, a service management platform has been a system of record. A user reports a problem, the platform logs it, routes it, tracks it against a service level, and reports on it afterwards. The intelligence and the action have always been human. Generative AI did not change that arrangement; it simply gave us faster note-takers assistants that could summarise a ticket or draft a reply while a person still did the work. Agentic AI is different in kind, not degree. An agentic system can interpret a request, decide what to do, and carry the task through to completion granting the access, resetting the credential, provisioning the device, closing the loop with a human supervising rather than executing.

This is the move from a system of record to a system of action, and it is the genuine breakthrough beneath the marketing. It is also why the word "agentic" is now stamped on virtually every product in the category, often with little behind it. The honest test for any IT leader is simple, and worth applying ruthlessly in every vendor conversation: does this system merely record and recommend work, or does it actually do the work, end to end, under controls I can see? Make vendors demonstrate the difference rather than describe it. A single live, audited resolution is worth more than any slide.

But the more important question is not whether the technology works. It is who controls it once it does. The instant you allow software to take action on your behalf rather than simply answer a question, governance stops being a compliance checkbox and becomes the core of the design. Three questions then matter more than any feature comparison.

Where does the data live, and where do the decisions happen? An agent that acts across your systems is reasoning over your most sensitive operational data. As Australian organisations particularly in government, financial services and other regulated sectors hand more authority to these systems, data residency and local accountability move from nice-to-have to non-negotiable.

Who is accountable when an agent acts? Autonomy without auditability is a liability. Every action an agent takes should be traceable, explainable and reversible, with clearly defined boundaries on what it may do without human approval. If a vendor cannot show you that audit trail, they are not ready for autonomous work, however impressive the demonstration.

And what happens if the platform changes hands? This is the uncomfortable lesson of the past year. When an independent AI capability is acquired by a larger platform, the customers who built on it can find their roadmap, their integrations and their pricing are suddenly someone else's decision. Capability you rent can be bought out from under you. That is not an argument against buying; it is an argument for understanding who ultimately controls the intelligence running your operations, and for valuing independence and local control as much as raw capability.

None of this should discourage adoption. The opportunity is real, and the mid-enterprise is in many ways better placed to seize it than the largest enterprises, whose scale makes change slow and expensive. A practical path exists. Begin with high-volume, well-bounded work where the right action is unambiguous password resets, access requests, routine provisioning and expand only as trust is earned through measured results. Design the human-in-the-loop deliberately, so that agents handle the toil and people apply judgement to the exceptions. Insist on auditability from the first day, not as a later add-on. And treat this as an operating-model change, not a software purchase, because that is what it is.

The temptation, in a year this loud, is either to wait for the noise to settle or to buy the boldest claim in the room. Neither serves you. The organisations that will benefit most are the ones that move deliberately: clear-eyed about what agentic AI genuinely changes, disciplined about governance, and thoughtful about who holds the controls. The service desk of the next decade will not merely record your organisation's work. It will do a great deal of it. The leaders who decide now, on their own terms, how much autonomy to grant and to whom, will be the ones still in control when it does.