While corporate investment in AI is projected to double in 2026, the industry focus has shifted toward more realism about how long it will take to see ROI. Meanwhile, tension is building as teams are strapped for talent and looking for solutions.
Deloitte’s State of AI 2026 report found that AI is delivering efficiency and productivity gains, yet revenue impact is still more of a goal than a result. While 74% of respondents hope AI will drive revenue growth, only 20% say it already is. This is reinforced by PwC’s Global CEO survey, which found that 56% of CEOs report neither higher revenues nor lower costs from AI so far.
What changes in 2026 is less about the investment and adoption we’ve seen up until now, and more about focus. Leaders want AI that helps this quarter, not more experimental fringe use cases. That puts pressure on IT organizations because they are the ones asked to keep the lights on while also moving AI forward. The trick is finding value without handing a probabilistic system the keys to critical infrastructure.
Workflows Already Exist, Agentic AI Changes the Trigger
The concept of a workflow in an automation tool isn’t new. ITSM and incident management platforms have supported workflows for years. For example, run these certain steps when an event occurs, send a webhook to another system, open a ticket, alert a team and wait for a manager to approve a change.
What’s new with agentic AI is the judgment around timing. Agents can look at what’s happening across a customer’s environment and decide when a workflow should run, then trigger it faster than a human can. With the rise of MCPs these workflows can now communicate more rapidly and easily with a larger breadth of systems.
According to the viral think piece ‘SaaS isn’t Dead,’ systems that control workflows will win in the agent era. “What wins: Workflow software that governs and captures decisions.” This is exactly where ITSM platforms come in. Organizations need to be in a position to capitalize on the rise of agents; ITSMs with strong workflow capabilities will provide that platform.
Operational delays are slowdowns in how IT operations respond, make decisions or execute work. These delays usually occur when teams are figuring out what’s actually going on, whether it’s real, how urgent it is and which playbook applies. Agentic systems can compress that determination step by correlating signals across tools and recommending or initiating specific workflow paths. Agentic AI can feel like a step change even though the underlying workflows look familiar.
The other change is around deciding who can build workflows in the first place. Historically, automation required specialized knowledge of the tool, its triggers, the scripts and all the ways things can break. Natural language interfaces are making it easier for teams to create and adjust workflows without being expert automation engineers. Governance is still critical, but it does make iteration faster.
Enterprise teams still have to draw a hard line between a trigger and taking autonomous action. The blast radius is not theoretical; real damage occurs if things go wrong.
The risk profile of infrastructure is different from most customer-facing AI use cases. A bad product recommendation might be annoying, but a bad infrastructure action can trigger an outage, expose data or compound across systems. The idea of autonomously adding memory sounds convenient until you consider reboots, licensing constraints, change windows and the very real possibility that memory wasn’t the problem.
Deloitte predicts that as the AI gap narrows, the next phase will be shaped by fundamentals like data hygiene, integration, governance and workflow fit instead of experimentation and demos. That is the reality check: IT leaders aren’t averse to AI agents, but they are averse to risk.
The Danger of Chasing Full Autonomy
Many IT leaders will restrict agentic systems to basic, bounded tasks in 2026. The reason is autonomous use can scale risk faster than it scales trust. When something breaks in the middle of the night, the question isn’t whether the agent followed instructions, but whether we can explain what happened, reverse it and verify that we remained compliant.
Enterprise environments tend to be more complex than they first appear. For example, an alert might point to a resource problem when the real cause is an upstream dependency or a bad deployment. Experienced operators learn to recognize these patterns over time, but AI agents today are not yet able to provide that judgment consistently and reliably.
If uptime, release safety and incident response speed are fundamental to a company’s competitive edge, it makes more sense to augment a team to protect that edge than to outsource to an autonomous black box.
AI is Actually Paying Off in ITSM
AI is showing real value inside IT operations and especially within IT service management (ITSM). There is a high volume of repetitive daily work that is often slowed down by context switching. These are back-office workflows that determine whether employees get help quickly or incidents spiral into outages. Even basic automation can create measurable improvements without increasing operational risk.
ITSM is basically an efficiency engine waiting for better tools. With the rise of MCP those tools are here.
Here are three use cases that stand out because they are measurable and relatively safe:
- Proactive incident detection: Fixing problems before they become tickets. AI can spot early warning signs by scanning monitoring data and past incidents so teams can respond sooner. This leads to faster detection, better prioritization and less downtime. It also gives IT clearer data to explain risk and system health to leadership
- Alert noise reduction: Saving humans from the barrage of alerts. Teams can burn out when too many alerts look urgent, even when they are not. AI can group related alerts, remove duplicates and add context so that engineers can focus on real problems. This cuts false alarms and speeds up response without letting AI take actions on its own. In ITSM, it also means fewer wasted escalations.
- Faster triage and routing: AI as a traffic cop instead of a judge. In ITSM, tickets can be routed incorrectly. AI can suggest the right owner, match issues with similar previous cases and pull the most relevant runbooks or knowledge articles. This value is quicker handoffs and fewer delays.
Deloitte noted that more workers now have access to approved AI tools, which makes these gains easier to scale if the data is clean.
The ROI Trap
Many IT organizations are going to be disappointed with their AI investments in 2026. In most cases, this will be the result of teams skipping foundational work. AI can’t clean up a messy knowledge base, fix undocumented processes or reconcile three conflicting answers to the same question.
The most common reason for failure is buying a tool before understanding where the bottleneck lies. An ITSM chatbot won’t reduce ticket volume if the knowledge articles are outdated. Automated categorization won’t help if the taxonomy is chaotic. In these instances, AI becomes a veneer covering broken workflows.
The other common reason for failure is measurement avoidance. Without a baseline, improvement can’t be recorded. It is important to define the problem, pick the metrics, then measure before and after.
Where Leaders Should Invest Right Now
Start with foundations that make every AI capability better:
- Clean up the knowledge base
- Enforce data hygiene so systems are trained on current information
- Pick one time-intensive workflow and fix it with AI intervention
- Invest in platforms instead of point solutions.
Governance and compliance aren’t disappearing with AI. If anything, they are more important than ever.
Prioritize AI that helps people rather than fully autonomous systems. Use it for suggestions, sorting and triage. These are things that save time without adding much risk. Letting AI change infrastructure automatically can backfire fast, and the fallout is costly. The goal is not to replace humans; it’s to cut the busywork so the team can focus on real problems.
