The next generation of AIOps calls for operational context

Knowledge can be shared. Context has to be built.

Over the last few months, we have built a line of reasoning about the evolution of modern IT operations. In the first series of articles, we discussed how the growing complexity of distributed environments profoundly changed the way we monitor systems, investigate incidents, and reduce response time.

We showed that monitoring more does not mean understanding better, that observability represents a new way of operating, and that reducing MTTR depends, above all, on the ability to quickly build a shared understanding of the incident.

In the second series, we turned our attention to Artificial Intelligence. First, we reflected on the fact that specialized agents do not, on their own, solve the main challenge of IT operations, in “AI agents do not solve the main problem of IT operations”. Then we argued that any agent depends on an element that comes before Artificial Intelligence itself: a shared operational context, capable of bringing together information from multiple sources, in “Before agents, there is operational context”. In the most recent article, we showed that specialization remains extremely valuable, but that an agent can know a technology in depth and still not understand an incident that crosses the entire architecture, in “An agent can know everything about a tool. Still, it may not understand the incident.”.

These reflections naturally lead to the question we intend to answer now.

If all agents will depend on a common context to produce consistent decisions, how will that context be built? And why will it become one of the most important assets of modern operations?

Answering these questions requires distinguishing two concepts that often appear mixed together in discussions about Artificial Intelligence, although they represent completely different challenges.

On one side is specialized knowledge. On the other, operational context.

At first glance, both seem to be part of the same problem. After all, both provide information for people and agents to make decisions. However, when we analyze how large organizations operate, we realize that they have completely different natures and, precisely for that reason, require equally different strategies.

Understanding this difference may be the most important step toward understanding why the next generation of AIOps will not be defined only by the evolution of Artificial Intelligence models, but by organizations’ ability to build and maintain a consistent representation of their own operation.

Specialized knowledge can be reused

One of the great advances provided by Artificial Intelligence is the ability to incorporate large volumes of specialized knowledge and make it extremely accessible. Today there are already agents capable of explaining how an Oracle database works, suggesting best practices for Kubernetes environments, interpreting metrics produced by observability platforms, supporting investigations in AWS environments, or guiding the configuration of tools such as ServiceNow and Dynatrace.

This movement tends to intensify.

As new models are trained and new knowledge bases are incorporated, we will see agents increasingly competent at answering technical questions, running specialized analyses, and supporting decisions in practically every discipline of IT operations.

The reason is simple: specialized knowledge has an extremely important characteristic, which is that it is widely reusable.

The best practices for administering an Oracle database do not change because the company belongs to the financial sector or to retail. The fundamental concepts of Kubernetes remain the same regardless of the country in which the organization operates. The official documentation of an AWS service remains valid for thousands of customers around the world, as do the recommended procedures for observability, network, or ITSM platforms.

Naturally, each organization will make adaptations, adopt different versions, or establish its own usage standards. Even so, the core of that knowledge remains essentially the same.

It is precisely this characteristic that makes specialization so compatible with Artificial Intelligence.

When knowledge can be documented, organized, and reused in different contexts, it becomes possible to train agents capable of reproducing that learning with enormous efficiency. The investment made to build that knowledge benefits thousands of organizations simultaneously.

This may be one of the greatest merits of specialized agents: they democratize access to the technical knowledge accumulated over decades of evolution of the technologies that sustain modern operations.

However, it is exactly at this point that an important limitation arises. Because what makes specialized knowledge so powerful, that is, its ability to be reused, is precisely what sets it apart from operational context.

Operational context cannot be reused

If specialized knowledge can be shared among thousands of organizations, why does that not happen with operational context?

The answer lies in the very nature of that context, because no company operates in exactly the same way as another.

Even organizations in the same sector, using similar technologies and facing similar business challenges, build different architectures over time. New applications are incorporated, integrations are developed, processes evolve, services are replaced, dependencies appear and disappear, changes are made continuously, and the environment comes to reflect the history of that organization.

It is precisely this history that turns operational context into a unique asset.

Consider two companies that use exactly the same Oracle database, run their applications on Kubernetes, store part of their infrastructure on AWS, and use the same ITSM platform. At first glance, we might imagine that an agent trained for this set of technologies would be enough to understand the incidents faced by both, but in practice, that hardly ever happens.

In one of them, a given application may depend directly on a centralized Oracle database. In the other, the same application may use a set of distributed microservices and different data synchronization mechanisms.

A change in an authentication service can affect completely different processes in each environment. The same observability alert can represent very different risks depending on the existing integrations, the business priorities, or the dependencies established over the evolution of that architecture.

Knowledge about Oracle remains exactly the same, but the operational context, however, becomes completely different.

This distinction helps explain why so many organizations discover that adopting specialized agents does not eliminate the need to build an integrated view of the operation. Agents can master the technologies used in depth. What they do not know, by definition, is the specific way those technologies were combined to sustain that company’s business.

In other words, they know the components, but they do not know the operation.

This difference is decisive because it moves the discussion about Artificial Intelligence to a more strategic level.

The main asset stops being only the ability to answer questions about known technologies and becomes the ability to faithfully represent the operational reality of a specific organization.

It is this representation that will make it possible to turn specialized knowledge into truly contextualized decisions.

Context becomes an asset of the organization

This change of perspective leads to an important consequence.

For many years, we treated technical knowledge as the main asset of IT operations. We invested in training specialists, documented procedures, created knowledge bases, and developed processes to preserve the experience accumulated by teams.

All of this effort remains indispensable, but the evolution of Artificial Intelligence introduces a new element into the equation. If specialized agents come to incorporate a good part of reusable technical knowledge, what will be the real competitive differentiator between two organizations that use similar technologies and have equally capable agents?

The answer will hardly lie in knowledge about Oracle, Kubernetes, AWS, or any other platform widely used by the market. That knowledge can be acquired, trained, and distributed at scale.

The differentiator will lie in what no other organization has: the representation of its own operational environment. This representation brings together information that does not appear in vendor documentation, is not available in technical manuals, and cannot be obtained simply by consulting a knowledge base. It embodies the relationships among applications, services, infrastructure, processes, changes, integrations, dependencies, and business priorities built over the evolution of that organization.

In other words, it represents the way the company really works, and that is precisely why operational context begins to take on the characteristics of a strategic asset.

Like customer data, intellectual property, or business processes, it becomes an asset that belongs exclusively to the organization. The more complete, up to date, and consistent that context is, the greater the ability of people and agents to quickly understand what happens during an incident, reducing uncertainty and supporting safer decisions.

This perception profoundly changes how we should think about investments in AIOps, and perhaps the most important question stops being “which agents should we adopt?” and becomes “how will we build and maintain the operational context that will allow all of them to work on the same understanding of the operation?”

The next generation of AIOps will be built on context platforms

If specialized knowledge can be learned, shared, and reused in practically any organization, while operational context has to be continuously built to reflect the specific reality of each environment, a consequence begins to become evident: modern operations come to require a new architectural layer.

For a long time, we invested in platforms responsible for monitoring infrastructure, collecting metrics, recording logs, tracking transactions, managing changes, and controlling incidents. Each of these solutions brought important gains in visibility and expanded our ability to understand different aspects of the operation. However, they were conceived to fulfill specific functions and, for that reason, continue to produce partial views of the environment. In the latest articles we discussed exactly this phenomenon.

The challenge of operations is no longer simply collecting information. It is also no longer just expanding observability or developing increasingly specialized agents. The real challenge became building an integrated understanding of the operation, connecting information that remains distributed across different platforms.

It is precisely from this need that a new layer of the architecture is born, and its function is not to replace existing tools or to compete with Artificial Intelligence agents. On the contrary. It exists to connect these two worlds.

This layer comes to organize the organization’s operational context, preserving the representation of the relationships among applications, infrastructure, services, integrations, dependencies, changes, and business processes. More than storing information, it keeps up to date a living model of the operation, able to continuously reflect the transformations the environment goes through.

When this context is available, specialized agents stop working on isolated interpretations and begin to share the same understanding of operational reality. An agent responsible for observability continues analyzing metrics. Another remains interpreting logs. A third continues tracking changes or supporting ITSM processes. The difference is that all of them develop their analyses on a common representation of the operation.

This detail profoundly changes the quality of decisions. Instead of multiple independent interpretations trying to explain the same incident, a shared context exists that makes it possible to relate those interpretations and turn them into a single understanding of the problem.

Perhaps this is precisely the main evolution of the next generation of AIOps. Not a generation defined only by more sophisticated Artificial Intelligence models, but by architectures capable of building, maintaining, and continuously making available the operational context on which those models will work.

When these agents leave isolated experiments and begin to operate in production on the same context, another complementary question arises: who will operate the agents when they reach production?

A new asset for operations

Throughout this series, we followed the transformation of IT operations from different perspectives. We started by discussing the difficulty of reducing operational complexity with new tools alone. Then we showed that monitoring more does not mean understanding better, that observability represents a new way of operating, that reducing MTTR depends on how fast we build understanding, and that specialized agents, although extremely valuable, remain limited by the partial view of each technology domain.

This article adds a new piece to that same reasoning. Specialized knowledge will remain indispensable. We will continue training specialists and using agents increasingly capable of interpreting specific technologies. This movement tends to accelerate and produce significant productivity gains for operations.

But the real competitive differentiator of organizations will not lie only in the quality of those specialists, human or artificial. It will lie in the quality of the operational context they manage to build. Because it is this context that turns knowledge into understanding. It is what makes it possible to connect information from multiple sources, preserve the relationships between architecture components, and offer a shared view of the operation to people and Artificial Intelligence agents.

Perhaps, a few years from now, we will look at operational context the same way we look today at strategic data, intellectual property, or business processes.

Not just as a set of information, but as an asset of the organization itself. And, if that really happens, a new question will arise naturally.

About the author

Paulo Florêncio de Menezes is an entrepreneur, partner, and co-founder of companies such as Target Solutions, BiPTT, and ARGUS.

He holds a degree in Data Processing from the Federal University of Amazonas (UFAM), an MBA in Business Management from Fundação Dom Cabral (FDC), and a Professional Master’s in Economics from Fundação Getulio Vargas (FGV).

In his Master’s research, he developed the IBEOP-IA (Brazilian Index of Potential Occupational Exposure to Artificial Intelligence), investigating the potential exposure of Brazilian occupations to AI capabilities by combining occupational data, artificial intelligence patents, and natural language processing techniques.