How To Redesign Infrastructure Operations For Agentic AI

Dilip Kumar is the President & Global Head of Infrastructure Solutions at NTT DATA.

getty​AI can operate at machine speed, but the operating models governing the infrastructure behind it often cannot. ​​

We saw this through our own AI transformation at NTT DATA. After we connected more than 21 enterprise data sources to AI systems, we still struggled to connect signals to context, context to decisions and decisions to governed execution.

An AI agent can identify an anomaly, retrieve information, invoke applications and recommend or initiate action within seconds. The infrastructure supporting that agent, however, often relies on fragmented tools, manual investigations, specialist handoffs and tickets sitting in queues. ​

Consider a familiar operational practice: An alert becomes a ticket and passes sequentially through specialist teams. AI may accelerate the diagnosis, but response times will remain slow if approvals and execution have to continue moving through the same queues.

For infrastructure leaders, solving this issue requires making three shifts.

Enterprise environments span multiple technology domains and providers. As a result, operational data, telemetry, security events and business context are often fragmented across platforms and teams.

A high-impact event rarely remains within one organizational boundary. A performance issue might appear as a poor user experience while its cause lies in the network, identity service, cloud platform or a security change. In these cases, one team sees the symptom, another holds the relevant telemetry and a third understands the business impact. Until those views are connected, teams are coordinating interpretations rather than resolving the problem.

In our environment, connecting operational, service, financial and workforce sources exposed relationships that individual tools could not. But even after aggregating data, we still had to answer four questions: What happened? Why does it matter? What should be prioritized? What action should follow?

Infrastructure leaders therefore need a shared operational layer across the multi-vendor estate, bringing together telemetry, service history, asset relationships and business context. Don’t think of it as another dashboard but as decision intelligence to reduce the time between detection, understanding and action.

​Tickets will not disappear. They remain important systems of record, providing auditability, governance and exception management. But they do not need to initiate and coordinate every infrastructure response.

The operating model changes when agents can correlate signals, identify probable causes, assess impact and take an approved action. The objective is bounded autonomy: Agents act where risk, authority and controls are understood, while people remain accountable for architecture, exceptions and judgment.

At NTT DATA, we developed this through a Client Zero project, where we tested our strategies first on our own operations before rolling it out to clients. We introduced specialist agents, coordinated execution and adopted a human + AI accountability model, which allowed AI to handle repeatable analysis and execution, people to manage exceptions and governance and leaders to remain accountable for outcomes.

We initially saw that faster diagnosis delivered limited value if a ticket still waited for assignment or approval. The larger improvement came when we redesigned the flow from incident to resolution and instead followed a loop of signal, sense, decide, act and improve.

In this case, across three delivery centers, mean time to resolution fell by 50%, triage accuracy reached 96.76% and AI agents handled 62% of tickets.

1. Gain context before AI. Agents require connected and reliable operational, service and business context, not simply more data.

2. Redesign workflows, not tasks. The biggest gains come from transforming end-to-end processes.

3. Prove outcomes before scaling. Adoption should be driven by measurable improvements in accuracy, resolution, effort, experience and risk.

4. Governance matters more as agent authority expands. Trust, accountability and control must be built in from the start.

In other words, agentic operations are an operating-model transformation, not simply a technology deployment.

Infrastructure operations have traditionally been measured through familiar metrics: tickets opened and closed, response times, availability and SLAs. Those measures remain important, but they provide an incomplete view.

We learned this by comparing activity measures with the value they were intended to create. A team can meet its SLA while issues recur, capacity remains underused or the business experience remains unchanged.

To solve this, we expanded what we measured in our Client Zero transformation. Alongside automation and productivity, we examined performance, effort, experience and prevention. Results included 30% less operating effort and customer satisfaction of 96%.

Instead of only asking in management conversations, “Did we meet the SLA?” leaders must also ask: Did infrastructure become more resilient? Was avoidable effort reduced? Did user experience improve? Did this prevent business impact?

This idea extends beyond NTT DATA or any one platform. Infrastructure is evolving from a collection of assets and support activities into a business capability measured by the outcomes it enables.​

AI is changing the speed at which infrastructure must operate. Technology foundations still matter, but success increasingly depends on how decisions are made, governed and measured.​ To better understand how AI-ready your infrastructure operating model is, here are four questions to ask:

1. Do you have a connected operational view? Can you understand an event across the full multi-vendor environment?

2. Can you turn data into a decision? Can you identify risks, dependencies, utilization opportunities and business impact quickly enough to act?

3. Have you defined the boundaries of agent authority? Which actions can agents execute independently? Which require human approval? How will those actions be governed and audited?

4. Are you measuring business outcomes? Are you improving business value or simply tracking operational activity? ​

Agentic AI’s reasoning capability will deliver little if the operating model can’t turn it into trusted action. By making these three shifts, organizations can be better positioned to move AI from experimentation into everyday operations, safely and at enterprise scale.​

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?