17 things business leaders don’t know about agentic AI in ITSM

Agentic AI is bringing tremendous change to industries by transforming the way businesses operate, automating human-driven workflows, taking authorized actions, and reducing the efforts of employees so they can focus on complex, high-value tasks. AI agents are quickly becoming part of the enterprise landscape, bringing agility and productivity to the organizations that use them.

This shift will affect managed service providers (MSPs) that provide IT Service Management (ITSM), the processes and tools organizations use to deliver, support, and maintain IT services, because they will need to agree with each client on how much autonomy agents can have and how agents will support the technology services people rely on. Those limits need to be clear before agents start handling work for the client.

For MSPs, agentic AI offers a way to support multiple clients more efficiently. Automated systems can help prioritize issues and document work across client environments, but agents that take action introduce additional responsibilities. Leaders must understand how delegating actions affects service delivery and accountability.

What agents can actually do

In many companies, team members call IT or submit a ticket when something goes wrong. There is a difference between IT answering support questions and agents authorized to resolve problems.

Here are 17 issues that business leaders need to understand when combining agentic AI with ITSM:

1. A conversational interface does not make a system agentic — People are used to talking to conversational interfaces, but traditional systems limit what they can do. They can provide information, but the user still has to take the next steps. An agentic system selects and carries out actions within authorized boundaries.
2. Autonomy has different levels — Clients give agents different degrees of independence. An agent might investigate, recommend a fix, request approval, or execute it. If the MSP provides the ITSM agentic system, business leaders must know which level they are purchasing or deploying.
3. Routine automation still has a place — Predictable tasks may need only routine automation rather than an agent. Reserve agents for work that requires interpreting context and choosing among approved actions.
4. A closed ticket does not prove the problem is fixed — A ticket can be marked complete even when the underlying problem persists. Measuring successful resolution means confirming that the service has recovered and monitoring for repeat incidents, rather than treating ticket closure as the outcome.

What must be ready before deployment

5. Poor service data can produce poor decisions — Outdated asset records, missing dependencies, and inaccurate configuration data can undermine an agent’s investigation.
6. The work extends beyond the service desk — Useful action may require connections to monitoring, identity management, endpoint tools, and client systems.
7. Each client needs its own operating boundaries — MSPs cannot assume one customer’s approved actions, maintenance windows, and escalation rules apply to another.
8. Permissions determine the potential damage — Limit agents to the systems and actions needed for their assigned work. Avoid broad administrative access by default.
9. Ticket content can become a security exposure — Instructions embedded in tickets, attachments, or retrieved material must not override approved operating rules.

Accountability when agents act

When something goes wrong, an agent can’t be held accountable the way a person can, so audit trails need to show who or what had authorization at each step.

Consider a cluster of endpoint alerts. The agent cross-references the alerts’ context with past incidents, decides on a course of action, and checks whether it has the authority to perform it.

If authorized, the agent takes action, making sure that the service is restored, and updates the ticket. Otherwise, the agent escalates the problem to an engineer, describing the circumstances and the course of action. The agent is responsible for investigating incidents and taking authorized actions, but liability for results should be shared between the MSP and the client and spelled out explicitly in the contract.

10. Delegating an action to agents does not transfer responsibility – Establish who owns the service outcome when an agent makes an incorrect change.
11. Approval needs clear triggers – Define which actions require review, especially those affecting production availability, security, or sensitive information.
12. An activity log must show more than the final answer – Record the information used, actions taken, approvals received, and results so teams can investigate failures.
13. Recovery needs to be planned before execution – Address rollback, stopping an agent, and handing the incident to an engineer with enough context to continue.

What changes for MSP economics and teams

14. Faster ticket handling is only one measure of value – Track service restoration, reopened incidents, escalation quality, SLA performance, and client satisfaction.
15. Agent activity has operating costs – Include model usage, integrations, monitoring, maintenance, and human review when evaluating savings.
16. Technical expertise remains essential – Engineers still investigate unfamiliar failures, maintain knowledge, and determine which actions can safely be delegated.
17. Pricing and contracts deserve another look – Examine how reduced manual effort affects service margins, pricing, and client expectations. Clarify permitted automated actions and responsibility for errors.

Choosing the right path

MSPs should offer agentic AI ITSM services, especially while some competitors still handle ITSM the way they did before AI. The service provider must understand that conventional automation, AI and agents can coexist within ITSM services. Conventional automation can handle predictable, rule-based tasks such as routing tickets or restarting a service under defined conditions. AI-assisted tools can analyze incident history, summarize findings, and recommend a response for an engineer to review.

But agentic AI goes further by interpreting the situation, selecting approved actions, and carrying them out within defined limits. An MSP does not need to replace every existing workflow with an autonomous agent. It can combine these capabilities according to the complexity of the task, the risk involved, and the level of human oversight required.

• Buy — If MSPs opt for a third-party platform with agentic AI capabilities, they may benefit from lower initial costs and faster deployment times. However, these products may not offer enough customization to meet the needs of the MSP and its clients. By purchasing a platform, the MSP or client may be locked into recurring licensing fees and vendor upgrades.
• Build — Another approach is to build a custom agentic ITSM platform that meets the business needs of the MSP and its clients. However, the MSP may not have the in-house skills to develop a proprietary platform. If that is the case, the MSP must find a software solution provider with agentic AI experience and industry-specific knowledge. This option potentially avoids ongoing third-party licensing fees because the business owns the intellectual property.
• Hybrid — If an MSP has an existing ITSM platform or if they choose to buy a new system, they can use an outside solution provider who can add agentic AI features.

Regardless of which path an MSP takes, it should establish a measurement framework before deploying agentic AI. Without a baseline, faster processing times or fewer manual steps do not necessarily demonstrate that the service has improved or that the investment is paying off.

Four categories of metrics can help MSPs and their clients evaluate results:

• Operational performance: Mean time to resolution, first-contact resolution, repeat incidents, and SLA compliance can show whether agents are helping teams resolve issues more effectively.
• Financial performance: Cost per resolved incident, human-review costs, platform and integration expenses, and net savings help establish whether the investment delivers a measurable financial return.
• Customer outcomes: Service availability, customer satisfaction, and business disruption help determine whether ITSM improvements translate into better service for the client.
• Risk and control: Unauthorized actions, failed remediations, policy violations, and recovery time help determine whether agents are operating safely and whether teams can recover quickly when an action goes wrong.

These measures should be evaluated together. An agent that closes tickets faster but increases repeat incidents or requires costly manual reviews may not deliver the expected value. Likewise, financial savings should not come at the expense of service availability or security. MSPs should compare results against pre-deployment baselines and review them regularly with clients to determine where greater autonomy is justified and where human oversight remains necessary.

Regardless of how MSPs develop AI and agentic IT services for their clients, the time is now.

Bredin, a boutique marketing research firm, found that 58% of small and midsized businesses planned to increase AI spending in 2026. However, many lack the staff to manage that investment. IDC, another research firm, reports that 40% of SMBs surveyed have no full-time IT employee.

These findings suggest an opportunity for MSPs to help clients deploy agentic AI and provide monitoring, maintenance and support.

Agentic AI gives MSPs a real opportunity to move beyond managing tickets and deliver more proactive, outcome-driven IT services. MSPs can build on this opportunity by continuously monitoring, improving and supporting AI-driven services. Ultimately, the real competitive advantage will come not from adopting agentic AI first, but from using it effectively to deliver better service, greater efficiency, and measurable business value.