Agentic AI in ITSM

back to list

Artificial Intelligence

0 Likes 11 mins read

Agentic AI in ITSM

1. Introduction: From Assistive AI to Agentic AI

2. What “Agentic” Actually Means in an ITSM Context

3. The State of Adoption in 2026

4. What the Major Platforms Shipped This Year

5. Ten Service Desk Workflows Agentic AI Is Already Handling

6. The Results So Far: MTTR, Deflection, and Productivity Gains

7. Where It’s Still Hard: Governance, Trust, and Skills Gaps

7.1 Trust is earned incrementally, not granted upfront

7.2 Auditability isn’t optional

7.3 Data quality is the quiet blocker

7.4 The skills gap is real and specific

8. Security and Risk: The New Non-Human Identity Problem

9. Future Outlook: 2027 and Beyond

9.1 Agentic spending will keep outpacing conversational AI spending

9.2 Multi-agent collaboration becomes normal, not novel

9.3 Interoperability standards like MCP will matter more than any single vendor’s roadmap

9.4 The role of the ITSM professional shifts from operator to supervisor and designer

9.5 Governance tooling becomes a product category of its own

9.6 Model flexibility, not model loyalty, becomes the competitive differentiator

10. Building a Roadmap: How to Adopt Agentic AI Without Getting Burned

10.1 Start with high-volume, low-risk workflows

10.2 Fix your data foundation before you scale autonomy

10.3 Build governance in from day one, not after an incident

10.4 Treat every agent as a new identity to secure

10.5 Invest in the specific skills gap, not a generic AI literacy program

10.6 Expand autonomy in proportion to demonstrated trust

11. Conclusion

1. Introduction: From Assistive AI to Agentic AI

For years, AI in ITSM was largely limited to assistive capabilities such as summarizing tickets, recommending knowledge articles, and drafting responses for human approval. While valuable, these AI copilots could advise but not act.

That is changing rapidly. Agentic AI is shifting ITSM from assistance to autonomy. Instead of simply recommending actions, AI agents can understand requests, plan resolution steps, execute tasks, and close tickets with minimal human intervention. A password reset is no longer suggested to an analyst; the agent completes it autonomously.

This represents a fundamental architectural evolution from single-step assistance to multi-step, outcome-driven workflows that operate across enterprise systems. As IT organizations pursue greater efficiency, resilience, and scalability, agentic AI is emerging as the foundation for the next generation of autonomous IT operations.

2. What “Agentic” Actually Means in an ITSM Context

There’s genuine debate in the ITSM community about what qualifies as “agentic” versus just well-orchestrated automation, and a lot of that debate hinges on whether a human was in the loop at any point. A reasonable working definition that’s emerged from 2026 vendor releases and analyst commentary looks like this:

Agentic AI in ITSM refers to systems that can reason about an incident, request, or change; plan a sequence of actions to resolve it; execute those actions across one or more connected systems; and adapt when something doesn’t go as planned — largely without human intervention, but within defined guardrails.

The distinguishing features tend to be:

Autonomy of execution: The system doesn’t just recommend a fix — it applies it: resetting a password, provisioning access, restarting a service, or updating a CMDB record.
Multi-step reasoning: It can chain several actions together (verify identity → check entitlement → provision access → notify requester → close ticket) rather than performing one isolated task.
Cross-system action: Modern agentic platforms increasingly reach beyond the ITSM tool itself into identity providers, cloud infrastructure, DevOps pipelines, and collaboration tools.
Governed autonomy: The credible platforms pair this with risk classification, permission boundaries, and audit trails — autonomy without governance is what turns a helpful agent into a liability.

That last point matters enormously, and we’ll come back to it, because it’s the single biggest differentiator between vendors that are shipping genuinely production-ready agentic ITSM and those that are still mostly repackaging chat.

3. The State of Adoption in 2026

The adoption data shows a market that has moved beyond AI experimentation but remains cautious about autonomous agents.

According to HCL Software’s State of Agentic AI in ITSM 2026 survey of 256 ITSM professionals, about three-quarters of organizations already use AI within ITSM, with service desks being the leading agentic AI use case. Among respondents able to assess outcomes, 94% reported measurable efficiency gains, indicating strong value realization. The study also found that organizations with AI in large-scale production are significantly more likely to pursue fully autonomous operations, highlighting that successful deployments drive greater interest in autonomy.

However, adoption challenges remain. The top barriers are data quality issues, governance concerns, and limited internal skills for managing agentic systems responsibly. Belitsoft’s 2026 forecast identified agentic AI as the fastest-growing enterprise technology priority, rising from 13.0% to 17.1% year over year, a 31.5% increase. Gartner further projects agentic AI spending to reach $201.9 billion in 2026, up 141% from 2025, with spending expected to surpass traditional chatbot and assistant investments by 2027.

Adoption also varies by region. ITSM.tools survey analysis highlights notable differences between North American and European organizations in both adoption speed and risk tolerance, reflecting the influence of regulations, data residency requirements, and organizational risk appetite on agentic AI deployment.

4. What the Major Platforms Shipped This Year

If 2025 was the year of agentic AI roadmaps and keynote demos, 2026 has been the year those roadmaps turned into shipped, licensed, generally available features across the major ITSM platforms.

ServiceNow made the most structurally significant move. In April 2026, it retired its long-standing five-tier licensing model (Standard, Professional, Enterprise, and their variants) in favor of three new AI-native tiers — Foundation, Advanced, and Prime — with legacy SKUs reaching end-of-sale on July 1, 2026. Crucially, AI capability is no longer an optional add-on: every tier now bundles an “Assist” usage allocation, unlimited Virtual Agent conversations, and a baseline of Now Assist and Predictive Intelligence out of the box. ServiceNow has also been rebranding its AI experience under the name “Otto,” with AI Agents built through a low/no-code AI Agent Studio that can reason and plan across connected systems (Azure, AWS, SAP via IntegrationHub), collaborate with other agents on cross-domain tasks, and — as of a July 2026 rollout — increasingly run on third-party model providers by default rather than a single proprietary model. A new capability called “Action Fabric” now lets external AI systems, including coding agents, trigger governed ServiceNow actions via the Model Context Protocol (MCP), effectively turning the platform into something other AI agents can call into directly. ServiceNow’s own leadership has been careful to frame this as augmentation rather than replacement: the company’s APJ innovation officer has publicly described the direction as “a collaborative partnership between humans and AI,” not full autonomy.

Ivanti made its own decisive move into commercial territory in April 2026 with an update to its Neurons platform introducing what it describes as an IT service desk AI agent — a system built specifically to handle incidents, requests, and knowledge search autonomously, without analyst involvement for a defined class of tickets. Industry coverage has framed this as one of the clearest signals that agentic ITSM has crossed from roadmap talk into commercial deployment: ticket deflection, long treated as an aspirational metric, is being described as an automated outcome rather than a target to chase.

HaloITSM, while earlier in its AI maturity curve relative to the hyperscale vendors, has been moving in the same direction — automatically categorizing and triaging incoming incidents and requests as a foundation for deeper agentic capability across its full process suite (incident, request, problem, change, and asset management).

Third-party and overlay agent platforms are gaining traction among organizations seeking faster AI adoption and cross-platform automation. Solutions like Enjo and Maven integrate with ITSM tools, enterprise knowledge sources, and collaboration platforms to resolve requests autonomously, escalate issues with full context, and streamline agent workflows. While native agents benefit from existing governance and security frameworks, overlay agents provide broader enterprise reach but add another layer of vendor management, security, and compliance oversight.

5. Ten Service Desk Workflows Agentic AI Is Already Handling

The theoretical promise of agentic ITSM is one thing; what it’s actually doing inside real service desks in 2026 is more concrete. The workflows that have proven safe and valuable enough to automate first share a common thread: high volume, well-defined logic, and low blast radius if something goes slightly wrong. These typically include:

Password resets and account unlocks — end-to-end, including identity verification against existing policy.
Standard access provisioning and deprovisioning — granting or revoking application access based on role and existing entitlement rules.
Ticket triage, categorization, and routing — reading an incoming request and assigning priority, category, and queue without human review.
Knowledge article retrieval and self-service resolution — resolving common “how do I” and known-error requests directly through conversational interfaces.
Software and hardware request fulfillment — provisioning standard catalog items (laptops, licenses, VPN access) automatically.
First-line diagnostic checks — pulling logs, checking system status, and running standard diagnostic scripts before a human ever sees the ticket.
Change risk scoring — flagging high-risk changes for additional review while auto-approving low-risk, well-precedented ones.
Proactive incident detection — correlating signals from observability tools to open (and sometimes resolve) incidents before an end user notices anything is wrong.
Status updates and requester communication — keeping requesters informed throughout a ticket’s lifecycle without manual follow-up.
Post-resolution ticket closure and documentation — summarizing what was done and closing the loop, including updating the knowledge base with anything learned.

Notably, industry framing around this list is explicit that the goal isn’t headcount reduction as an end in itself — it’s freeing human specialists from repetitive, low-judgment work so they can focus on the incidents and changes that genuinely need expert judgment, negotiation, or creativity.

6. The Results So Far: MTTR, Deflection, and Productivity Gains

The results from mature 2026 agentic AI deployments help explain growing enterprise interest. Advanced implementations report auto-resolution rates of up to 84% for IT support requests, while pilots have reduced Level 1 and Level 2 ticket volumes by 60–80%. Among leading adopters, mean time to resolution has fallen by more than half, from about 51 hours to 23 hours.

These gains signal more than operational efficiency. As routine issues are resolved automatically within minutes instead of remaining in queues for days, service desk teams can shift their focus from repetitive troubleshooting to managing exceptions, complex changes, and overseeing AI agent ecosystems. However, these figures mainly reflect highly mature deployments. Many organizations remain in pilot phases, meaning the substantial benefits seen by advanced adopters should not be viewed as typical outcomes for first-year implementations.

7. Where It’s Still Hard: Governance, Trust, and Skills Gaps

For all the momentum, the honest state of the market in 2026 is that agentic AI in ITSM remains genuinely difficult to govern well, and the surveys reflect that directly. Data quality issues, governance concerns, and a shortage of internal skills to deploy agentic systems responsibly consistently top the list of adoption barriers reported by ITSM professionals.

A few specific governance challenges keep coming up across vendor and analyst commentary:

7.1 Trust is earned incrementally, not granted upfront

Adoption research shows a strong relationship between how much organizations trust their AI systems and how far along their adoption journey they are — trust tends to build after early wins, not before them, which argues for starting with low-risk, high-volume use cases rather than the flashiest possible autonomous workflow.

7.2 Auditability isn’t optional

Analyst commentary on where agentic ITSM is headed emphasizes a future built around role-based AI specialists that operate strictly within defined workflows, respect existing permission structures, and produce a full audit trail for every action taken — not just the outcome, but the reasoning path that led there.

7.3 Data quality is the quiet blocker

An agent that reasons well off bad or stale CMDB, knowledge base, or entitlement data will confidently execute the wrong action. Several analysts describe this as where a large share of predictive and agentic ITSM projects actually die — not in the AI model, but in the underlying data foundation it’s reasoning over.

7.4 The skills gap is real and specific

It’s not a generic “our team doesn’t understand AI” problem — organizations increasingly report needing staff who understand how to build, license, and govern agentic workflows specifically, distinct from general ITSM administration skills. That specialization is already showing up as a distinct line item in job postings and compensation data for platform-specific roles.

8. Security and Risk: The New Non-Human Identity Problem

Agentic AI introduces security risks that traditional assistive AI never faced because agents do more than generate content, they execute actions on enterprise systems. Industry analysts warn that autonomous agents represent a distinct risk category, with compromised agents potentially exposing sensitive customer data, altering business processes, or impacting critical production systems.

This shift is redefining identity and access management. By 2026, organizations are expected to manage significantly more machine and agent identities than human users, while existing IAM frameworks remain largely human-centric. At the same time, prompt injection and agent-targeted attacks are projected to increase, amplifying the financial and operational impact of autonomous system misuse.

For ITSM leaders, every AI agent should be treated as a privileged non-human identity. Agent governance requires strict least-privilege access, credential rotation, and continuous behavioral monitoring, especially since autonomous agents can operate at machine speed and scale, making security lapses far more consequential.

9. Future Outlook: 2027 and Beyond

9.1 Agentic spending will keep outpacing conversational AI spending

Gartner’s projection that agentic AI spending overtakes chatbot and assistant spending by 2027 suggests the market center of gravity is shifting decisively toward autonomous execution rather than conversational assistance, even for vendors that built their reputation on chat interfaces.

9.2 Multi-agent collaboration becomes normal, not novel

ServiceNow’s own AI Agent Studio already supports agents collaborating with each other across domains rather than operating as single, siloed bots. Expect this to become the default architecture: an incident-response agent, a change-risk agent, and an identity-provisioning agent working together on a single complex request rather than one monolithic “IT agent” trying to do everything.

9.3 Interoperability standards like MCP will matter more than any single vendor’s roadmap

ServiceNow’s move to let external agents — including coding assistants — trigger governed platform actions through the Model Context Protocol points toward an ITSM future where the interesting boundary isn’t “which vendor’s AI” but “which agents, from whichever vendor, can safely act on which systems, under what governance.” Organizations that architect for this kind of interoperability now will have far more flexibility than those who lock themselves into a single vendor’s closed agent ecosystem.

9.4 The role of the ITSM professional shifts from operator to supervisor and designer

Broader agentic AI forecasting suggests that by 2030, a large majority of technical staff will spend more time planning, reviewing, and governing autonomous systems than performing the underlying tasks those systems now handle directly. For ITSM specifically, this likely means fewer people manually triaging tickets and more people designing agent workflows, tuning risk thresholds, and auditing agent decisions.

9.5 Governance tooling becomes a product category of its own

Expect continued growth in dedicated AI governance layers — the kind of centralized control-tower functionality ServiceNow and others are building — as organizations realize that agentic autonomy without centralized oversight isn’t a viable long-term operating model, even if it looks efficient in a pilot.

9.6 Model flexibility, not model loyalty, becomes the competitive differentiator

As frontier model performance continues to converge across vendors, the decisions that will actually matter for ITSM buyers are architectural: can you route between models for cost and accuracy, and are you locked into one provider’s API in a way that limits your options later. The platforms winning trust in this space are increasingly the ones offering model flexibility rather than betting everything on one proprietary model.

10. Building a Roadmap: How to Adopt Agentic AI Without Getting Burned

For technology leaders evaluating where to start, the pattern that’s emerged from 2026’s successes and near-misses points toward a fairly consistent sequence:

10.1 Start with high-volume, low-risk workflows

Password resets, standard provisioning, and knowledge-based self-service resolution are the proven entry points precisely because the cost of an occasional mistake is low and recoverable.

10.2 Fix your data foundation before you scale autonomy

An agent is only as good as the CMDB, knowledge base, and entitlement data it reasons over. Treat data quality remediation as a prerequisite, not a parallel workstream.

10.3 Build governance in from day one, not after an incident

Risk classification, human-in-the-loop checkpoints for higher-risk actions, and complete audit trails should be part of the initial design, not a bolt-on after something goes wrong.

10.4 Treat every agent as a new identity to secure

Apply the same least-privilege, monitoring, and credential-management discipline to agent accounts that you would to any privileged human account — arguably more, given the speed and scale at which agents can act.

10.5 Invest in the specific skills gap, not a generic AI literacy program

The people who will make agentic ITSM work are the ones who understand how a specific platform’s agents are built, licensed, and governed — that’s a narrower and more technical skill set than general AI awareness training.

10.6 Expand autonomy in proportion to demonstrated trust

Adoption data consistently shows trust and autonomy scaling together over time. Resist the pressure to leap straight to fully autonomous, high-risk workflows before your organization — and your agents — have earned that trust through smaller wins.

11. Conclusion

Agentic AI in ITSM has moved decisively out of the pilot phase in 2026. The major platforms have shipped real, licensed, production capability; adoption is broad and growing; and the efficiency numbers from mature deployments are genuinely impressive. But the same data that shows this momentum also shows an industry still working through the harder problems — governance, data quality, non-human identity security, and the skills needed to run all of it responsibly.

The organizations getting the most value aren’t necessarily the ones moving fastest toward full autonomy. They’re the ones treating agentic capability as something to be earned incrementally: strong data foundations, governed permissions, audit trails from the start, and autonomy that expands only as trust is demonstrated. That’s a less dramatic story than “AI runs your service desk now” — but it’s the one the 2026 evidence actually supports, and it’s the version of this technology that’s likely to still be standing, and still trusted, by 2030.

Reference

Likes

Author Details

Vidya Anandrao Jadhav

Vidya is a senior consultant handling research and is responsible for delivering client requirements through the iCETs unit of Infosys. She holds considerable experience in catering to the research requirements for multiple domains.