In Service of Infrastructure: How Multi-Agent Systems Make IT Operations Manageable
Monolithic IT infrastructures are a thing of the past. Growing complexity is pushing traditional operating models in IT operations to their limits.
Guest commentaryOliver Köth, Managing Director Technology and Innovation DACH at NTT DATA · Frank Beckereit, Senior Director Consulting at NTT DATA
Monolithic IT infrastructures are a thing of the past. Today, connected applications, cloud services and platforms shape IT operations. Growing complexity is increasingly pushing traditional operating models to their limits. The good news: multi-agent systems already exist as solutions that automate entire IT process chains in a targeted way and help people identify correlations and causes more quickly.
Key takeaways
- Specialized agents instead of a single central system. Separate agents evaluate logs, metrics, configuration states and alerts. An orchestration layer decides which agent is called in when.
- The bottleneck is correlation. Automated classification of operational data is established. Correlating information across systems remains the unresolved challenge.
- Phased implementation. First analyze and prioritize, then execute low-risk operations automatically. Governance rules define where escalation to humans occurs.
The initial trigger often seems unremarkable: an alert from the network, an unusual application response time or a failed service. What follows in many IT departments, however, is often a laborious search for the needle in the haystack, namely the root cause. Is the problem in the network? In an API? In the data center? Or with a cloud service? With every additional application, every new interface and every outsourced service, the dependencies between the individual components grow. As a result, cause and effect are located in the same place less and less often.
Conventional monitoring and operations tools reliably capture the state of individual systems, yet they increasingly reach their limits when it comes to interpreting these relationships and their impact on day-to-day business. The reason lies less in the complexity of modern IT landscapes than in the volume of log and operational data, which has grown steadily for years. Only once this information has been classified and correlated across systems can correlations and causes be reliably identified. While automated classification is now well established, AI-powered intelligent correlation marks the next evolutionary step. How does this work?
Unlike conventional approaches, multi-agent systems rely on several specialized agents that take on different tasks and share their findings with each other. While some agents analyze logs, evaluate metrics or assess network events, others bring this information together and place it in a common context. Step by step, this produces an overall picture that makes correlations between distributed systems visible and helps IT teams identify the causes and effects of incidents far more quickly.
IT needs new architectures
Multi-agent systems model IT operations as a distributed processing system. Instead of relying on a central component to analyze all data, tasks are distributed among specialized agents, each aligned with a clearly defined functional area. These agents work continuously in the background, always placing incoming event data into the current system context.
Typical specializations include analyzing log data, assessing configuration and device states, classifying alerts or correlating events across multiple systems. Other agents take on tasks such as identifying historical patterns in operational data or recognizing recurring failure modes. The major advantage over conventional architecture approaches lies in the automated, orchestrated collaboration of the functional units.
One level above, a central orchestration layer coordinates the activation and interaction of the agents. It determines which agents are brought in for which situation, how results are merged and in which context decisions are prepared. This structure is complemented by governance rules that define what the agents are permitted to do. These rules specify which agent may execute which actions automatically and when an issue must be escalated to the people responsible. The result is a system that distributes operational tasks efficiently and retains human control over critical decisions.
Step-by-step transformation

The introduction of agentic systems in IT operations usually proceeds step by step and builds on existing automation and monitoring structures. In the first stage, agents take on supporting functions by analyzing data, prioritizing alerts and providing contextualized information, while the actual decision remains with the human operations team. The added value here comes primarily from preprocessing large volumes of data and from sharply reducing the information noise that would otherwise grow ever louder.
In the next stage of development, agents can automatically execute clearly defined, low-risk operations. These include standardized responses to known failure scenarios or recurring configuration problems. These interventions take place within defined boundaries and are systematically logged, so their effects remain traceable and can feed into future decisions. The systems are thus already moving toward active operational support while always remaining under full control.
In the next stage of development, this approach can enable an operating model in which agent systems handle more than small-scale scenarios. They then also actively establish correlations between events on a large scale and automatically derive concrete measures from them. Several agents are brought in to analyze causes, evaluate hypotheses and prioritize possible courses of action. Decision-making continues to follow defined governance rules that ensure critical interventions remain traceable and can be escalated when necessary.
Structural impact on IT operations
With the introduction of agent-based architectures, the operational reality of IT changes fundamentally. Large volumes of events are filtered, correlated and assessed in the early phases of processing, so that only a significantly reduced share reaches day-to-day operations as genuinely relevant incidents. The focus of work shifts from handling individual notifications to interpreting relationships across the system.
This development also affects how roles are distributed within IT organizations. While agents take over repetitive, easily automated operational tasks, tasks such as defining automation rules, monitoring agent systems, assessing relevant events and continuously refining operational logic become more prominent. IT operations thus becomes more of a control function for distributed systems. The focus then lies on the stability and efficiency of the overall system rather than the individual ticket.
All of this relieves IT teams and also enables a new level of operational control. Continuously correlating events across system boundaries allows patterns that often remain hidden in conventional monitoring architectures to be identified more quickly, or identified for the first time. This is especially true of complex failure chains that unfold across multiple systems and can only be explained when viewed as a whole. With this development, the boundaries between monitoring, analysis and operational execution also blur. IT infrastructures are evolving into systems that continuously monitor and interpret their own states and act on them within defined boundaries. The focus thus shifts from merely reacting to events toward structured control of complex system states.
Oliver Köth is Managing Director Technology and Innovation DACH at NTT DATA.
Frank Beckereit is Senior Director Consulting at NTT DATA.
Frequently asked questions
How do multi-agent systems differ from conventional monitoring?
Conventional tools reliably capture the state of individual systems. Multi-agent systems distribute the analysis across specialized agents that merge their findings and place them in a shared context.
Where is the real bottleneck in IT operations?
Automated classification of operational data is now largely established. What remains open is correlation across systems, which makes the interdependencies between distributed systems visible.
How do companies introduce such systems?
In stages. First, agents analyze data and prioritize alerts while the operations team makes the decisions. After that, they automatically execute clearly defined, low-risk operations.
Who keeps control over critical interventions?
Governance rules define which actions each agent may perform and when an issue must be escalated to the people responsible. Interventions are logged and remain traceable.
Editor's Picks
cloudmagazinObservability quietly eats the cloud budgetcloudmagazinWhen AI agents travel: data residency as an operating problemcloudmagazinObservability: More than monitoring with logs and metricsMore from the MBF Media Network
MyBusinessFutureWindows 11 limits what AI agents can accessDigital ChiefsCIOs Should Make AI Learning Time a Fixed Part of Working HoursSecurityTodayBlack October: Many attacks had been running for monthsImage source: AI-generated (August 2026), in-article image GettyImages/NTT DATA
Translated from the German original using artificial intelligence. The German version is authoritative.

