Agentic service desk: what AI agents mean for ITSM by 2028

The short answer: The IT service desk is one of the first places where AI agents are moving from pilots into daily work. In a Gartner survey of 782 infrastructure and operations (I&O) leaders, published on 7 April 2026, "53% of I&O leaders reported their AI wins occur in ITSM" (Gartner, 7 Apr 2026). The organisations that benefit will fix their knowledge, data and permissions first, then raise agent autonomy step by step under clear governance.

This guide is for IT directors, heads of IT operations and ITSM platform owners in the UAE and GCC, whichever ITSM platform you run (ServiceNow or another).

Why the service desk is an early mover

IT is where agent adoption is most visible. In McKinsey's 2026 State of AI survey, respondents most often report scaling AI agents within IT, knowledge management and software engineering. The same survey finds cost reductions from AI are most frequently reported in supply chain management, service operations and manufacturing (McKinsey).

The reasons are practical. Service desks already hold large volumes of structured tickets, a catalogue of standard requests and a knowledge base. Many requests follow repeatable patterns. And the people who would supervise agents, IT staff, are close to the technology.

What Gartner's public research says about IT operations

Gartner's newsroom gives a balanced picture: real wins, a high failure rate and a long road to full autonomy.

Results are mixed. "Only 28% of AI use cases in infrastructure and operations (I&O) fully succeed and meet ROI expectations, while 20% fail outright," according to a Gartner survey of 782 I&O leaders in November and December 2025 (Gartner, 7 Apr 2026).
Ambitious autonomy fails most often. "I&O leaders most frequently observe AI failures in auto-remediation, self-healing infrastructure, and agent-led management of workflows within and between systems" (Gartner, 7 Apr 2026).
Data and skills decide outcomes. "38% of I&O leaders said poor data quality or limited data availability was a direct cause of AI project failure", and 38% of those who faced setbacks cited persistent skill gaps (Gartner, 7 Apr 2026).
Agents are coming to the tools. At its Sydney I&O conference, Gartner stated: "Gartner predicts AI agents will be implemented in 60% of all IT operations tools by 2028" (Gartner, 11 May 2026).
Full autonomy is further away. The same update says: "Gartner analysts predict only 15% of agents will be fully autonomous within the next seven years" (Gartner, 11 May 2026).

Gartner's advice from that keynote fits the service desk well: "The best way for I&O leaders to get started with agentic AI is with practical, foundational use cases in IT operations" (Gartner, 11 May 2026).

These are survey results and forecasts, not certainties. Taken together, they point to a service desk where agents take on more routine work under human supervision, rather than one that runs itself.

What changes on the desk

Today	With agents (gradually)
Analysts read, categorise and route every ticket	Agents triage and route; analysts review exceptions
Knowledge articles are written after the fact, if at all	Agents draft articles from resolved tickets for human approval
Standard requests wait in a queue	Agents complete approved standard requests within set limits
Major incidents rely on manual correlation	Agents summarise signals and suggest likely causes for engineers to confirm
Success is measured by tickets closed per analyst	Success includes agent accuracy, exception rates and cost per resolution

Each row is a choice, not an automatic outcome. Decide which tasks agents should observe, advise on or act on, and in what order.

Fix the foundations first

Agents amplify whatever they are given. If the knowledge base is out of date or the configuration data is wrong, an agent will act on bad information faster than a person would.

Knowledge

Remove or update outdated articles, and set an owner and review date for each.
Write articles in a consistent structure so that agents can use them reliably.
Close the loop: every resolved ticket should either use an article or show that one is missing.

Configuration and service data

Check the accuracy of your configuration management database (CMDB) for the services in scope.
Make sure each service has a clear owner, support group and escalation path.

Catalogue and workflows

Simplify the service catalogue. Ambiguous request types confuse people and agents alike.
Document approval rules explicitly, so that an agent can follow them and an auditor can check them.

Identity and access

Give each agent its own identity, not a shared admin account.
Grant the minimum permissions needed, and log every action an agent takes.

Govern agents by what they are allowed to do

Applying the same controls to every agent causes problems in both directions. In May 2026, Gartner predicted that by 2027, 40% of enterprises will demote or decommission autonomous AI agents because of governance gaps identified only after production incidents. It recommends classifying agents by autonomy level, with controls that grow at each level (Gartner). On a service desk, that might look like this:

Observe: an agent searches knowledge and summarises ticket history for the analyst. Controls: scoped data access, authentication, usage logging.
Advise: an agent suggests a category, priority or resolution, and the analyst decides. Controls: accuracy testing and guidance on how far to rely on suggestions.
Act with approval: an agent prepares a change, such as resetting access or updating a record, and a person approves each action. Controls: approval workflows with audit trails and agent-specific incident procedures.
Act autonomously: an agent completes defined standard requests without approval, and people review exceptions and outcomes. Controls: continuous monitoring, rollback and automatic stops when thresholds are breached.

Most organisations should expect to spend some time at levels 2 and 3 before allowing autonomy on anything that changes production systems.

Measure agents like any other service

Treat agent performance as something you specify, measure and, where vendors are involved, write into contracts. Before any agent goes live, define:

Accuracy: how often the agent's categorisation, routing or resolution is correct, by ticket type.
Exception rate: how often it hands work back to people, and why.
Time and cost: time to resolution and the running cost per resolved ticket, including AI usage charges.
Safety: incidents caused or worsened by agent actions, and time to roll back.

Report these alongside your existing service levels, so that the desk is managed as one system of people and agents.

Plan for people, not just technology

So far, expectations about job losses have run ahead of reality. In McKinsey's 2026 survey, 14% of respondents from organisations using AI report that AI contributed to an overall decline in workforce size in the past year, less than half the 32% who expected reductions in the previous year's survey. Looking ahead, the largest share of respondents expects reductions within service operations and supply chain management (McKinsey, 25 Aug 2026). For service desks, the more useful question is how roles change. New responsibilities include supervising agent queues, maintaining knowledge, tuning workflows and investigating agent errors. Plan training and career paths for these roles early, and involve your service desk team in designing the new way of working.

A phased approach

Assess: measure knowledge quality, CMDB accuracy and ticket patterns for the services in scope.
Pilot: run agents in observe and advise modes on a small set of high-volume ticket types.
Measure: compare accuracy, resolution times and costs with the current process.
Expand carefully: allow act-with-approval for low-risk standard requests, with full logging.
Review: consider autonomy only where the evidence supports it, and keep a clear off switch.

The bottom line

The agentic service desk is coming, but it will be built on ordinary foundations: accurate knowledge, clean service data, sensible permissions and honest metrics. Leaders who invest in those now will be able to raise agent autonomy with confidence, rather than rolling it back after an incident.

Sources

McKinsey, "The state of AI in 2026: On the road to ROI" (25 Aug 2026). https://www.mckinsey.com/capabilities/quantumblack/our-insights/the-state-of-ai
Gartner, "Gartner Says AI Projects in I&O Stall Ahead of Meaningful ROI Returns", Q&A with Melanie Freeze (7 Apr 2026). https://www.gartner.com/en/newsroom/press-releases/2026-04-07-gartner-says-artificial-intelligence-projects-in-infrastructure-and-operations-stall-ahead-of-meaningful-roi-returns
Gartner, "Gartner IT Infrastructure, Operations & Cloud Strategies Conference 2026 Sydney: Day 1 Highlights" (11 May 2026). https://www.gartner.com/en/newsroom/press-releases/2026-05-11-gartner-it-infrastructure-operations-and-cloud-strategies-conference-2026-sydney-day-1-highlights
Gartner, "Gartner Says Applying Uniform Governance Across AI Agents Will Lead to Enterprise AI Agent Failure" (26 May 2026). https://www.gartner.com/en/newsroom/press-releases/2026-05-26-gartner-says-applying-uniform-governance-across-ai-agents-will-lead-to-enterprise-ai-agent-failure