Putting AI in the process requires redesigning the process

A company can incorporate artificial intelligence into practically every step of a process and still keep running almost the same process as before. The difference is that some tasks are now performed faster.

One agent looks up information that someone used to search for manually. Another prepares an analysis, fills in a system, or forwards a request. A third follows a queue and suggests the next action. Individually, each application can produce real productivity gains, but the flow still carries the same steps, approvals, handoffs of responsibility, and dependencies that existed when all the work was done by people.

This distinction is becoming important because corporate AI adoption is moving faster than the redesign of how companies work. In the global survey published by McKinsey in 2025, only 21% of respondents using generative AI said their organizations had fundamentally redesigned at least some workflows. When comparing 25 attributes related to adoption, workflow redesign was precisely the one with the greatest effect on the likelihood of the organization seeing EBIT impact from AI [1].

The problem, therefore, begins to change. After learning to build copilots and agents, organizations need to decide whether they will keep fitting these technologies into the current way of working or use their new capabilities to rethink the process itself.

Automating a step is different from redesigning the flow

The first approach is naturally simpler. We take a known process, identify time-consuming tasks, and introduce AI where it can help.

Imagine an incident handling process. An agent can summarize alarms, consult documentation, look for previous occurrences, and prepare a recommendation for the specialist. Each of those activities can save a few minutes and improve the productivity of whoever is investigating the problem.

But the incident may keep following exactly the same path. One system generates the alert, someone does the triage, another professional looks up information in different tools, a third team needs to be called in, certain decisions wait for approval, and the knowledge produced during the investigation does not always remain available for the next occurrence.

Putting AI in some of those steps improves the existing process. Redesigning it requires a different question: if people and agents could work together from the start, would this process still need to work this way?

The difference seems small, but it changes the starting point. Instead of looking for where to fit AI, we start looking at the result we want to produce, the decisions needed to get there, the information each decision requires, and which activities make more sense for people, traditional automations, or agents.

When we use AI only to speed up existing steps, we may make an old process faster without necessarily making it better.

This discussion appears quite clearly in the most recent research. In 2026, McKinsey returned to the topic when analyzing companies that were managing to extract more value from AI. Among the so-called top performers, organizations that attributed at least 5% of their EBIT to the technology, the likelihood of carrying out a broad redesign of the operating model was three times higher, and the likelihood of redesigning workflows before selecting AI tools was twice as high [2].

The order of decisions matters. When we start with the tool, it is natural to look for tasks it can perform. When we start with the process, we may discover that some of those tasks should not even continue to exist.

Agents change what can be considered a process

The need for redesign becomes even more evident when we move from copilots into the universe of agents.

A copilot normally helps someone carry out an activity. An agent can receive a goal, query different sources, choose tools, execute a sequence of actions, and decide which path to follow according to what it finds. In more advanced architectures, several agents can collaborate with one another and coordinate different parts of the same flow.

This creates possibilities that simply did not exist when many of today’s corporate processes were designed.

Consider incident handling again. Instead of just summarizing alarms for an operator, agents could gather signals from different platforms, check recent changes, retrieve knowledge about previous occurrences, identify known dependencies, and prepare a hypothesis about what is happening. Depending on the rules and the level of autonomy allowed, some actions could be executed automatically, while others would continue to require human validation.

In this scenario, the discussion is no longer only about which tasks can be automated. We start discussing how work should flow between people, agents, automations, and systems.

This is what Deloitte calls agent-first process redesign. In its analysis of strategy for agentic AI, the consultancy argues that more advanced organizations do not simply add agents to existing workflows; they look at processes end to end and redesign them to take advantage of specific capabilities of agents [3].

A later Deloitte survey shows how big the distance is between that vision and current reality. Only 5% of the organizations surveyed said their processes were highly prepared for AI agents, while only 15% had already scaled orchestrated multi-agent adoption across different functions. At the same time, 74% expect that, within four years, at least half of their processes will have been redesigned or rebuilt around agents [4].

The transformation starts to get deeper when we stop asking “where do we put an agent?” and start asking “how should this work happen now that agents exist?”.

This shift also helps avoid a recurring automation trap: preserving unnecessary complexity because it is already part of the process. A flow created over years can accumulate checks, handoffs between areas, and manual steps that appeared to compensate for technological or organizational limitations of another era. Automating all of that does not eliminate the complexity; in some cases, it only allows it to happen faster.

AI opens room to review those assumptions, but it does not do that work on its own.

Redesign also redefines the role of people

Rethinking processes for AI does not mean designing them to take people out of the way. In many cases, it means deciding more clearly where human participation really produces value.

There are activities where speed and scale matter more. Others require judgment, accountability, negotiation, tacit knowledge, or an understanding of consequences that should not be fully delegated. Between these extremes there is a large amount of work in which people and agents can act in a complementary way.

That is why process redesign cannot be only technical. It involves responsibilities, decision criteria, supervision points, exception handling, and an explicit definition of the limits of autonomy.

Deloitte itself found that 75% of the executives surveyed believe that collaboration between people and agents creates more value than automation based solely on agents [4]. The question, therefore, is not simply determining how much work can be transferred to AI, but designing a division of responsibilities that makes sense for each process.

Redesigning a process for AI also means redesigning the relationship between autonomy and human oversight.

This point connects directly to the discussions we developed in previous articles. The greater the autonomy granted to agents, the more important context, observability, governance, and the ability to intervene become. A redesigned process needs to consider these conditions from the start, and not add them after the automation is already running.

In an IT operation, for example, governance and operation of agents need to be present when an agent investigates a wide set of signals without human intervention, correlates information, and suggests a probable cause. Certain low-risk actions can be authorized automatically, while changes with greater impact continue to depend on approval. If the agent finds a situation for which it lacks sufficient context or adequate confidence, the flow itself can provide for handing off to a specialist.

The human role thus stops being defined only by the task the person performed before AI and starts to be defined also by the responsibility that needs to remain human within the new process.

This may be one of the reasons why transforming processes is harder than deploying tools. Buying technology is a relatively bounded decision. Changing how decisions are made, how areas interact, and who answers for each result touches structures that were built over years.

The advantage will not come from having more agents

As agents become easier to create and are incorporated into the enterprise platforms themselves, having agents will quickly stop being a differentiator.

The difference will lie in the ability to use them within processes that have been designed to combine technology, information, and people in a coherent way.

McKinsey describes this evolution in three horizons. In the first, AI expands individual capabilities; in the second, it starts to automate end-to-end workflows; in the third, the organization stops starting from the existing process and begins to redesign work, roles, and the operating model with AI as a structural part of the new configuration [5]. In the same survey, leaders of organizations that redesigned workflows were 5.3 times more likely to report capturing business value than those that kept flows unchanged [5].

This does not mean every company needs to redesign every process at once. It means recognizing that there is a limit to the value obtained when a new technology is forced to work permanently within an old logic.

The greatest gain from AI may not be in running faster the process we already know, but in finally allowing us to ask why that process still works this way.

For many organizations, this will be a gradual transition. First come assistants and copilots. Then agents take on more complete tasks. Next, different agents begin to take part in the same workflow. At some point, keeping all the previous steps and boundaries makes less and less sense.

At that stage, the discussion about AI no longer belongs only to technology. It starts to involve the design of the organization itself.

From vision to practice: ARGUS

When AI agents start to take part in an organization’s processes, an opportunity arises that goes beyond automating existing tasks. We can rethink how information is shared, how decisions are made, and how responsibilities are distributed among people, systems, and agents.

In IT and Telecom operations, this challenge is particularly evident. Important information remains spread across monitoring, observability, inventory, changes, and ITSM. ARGUS connects these different sources and builds a shared operational context, reducing the need to manually rebuild the situation at every incident or decision.

With AgentOps and AI Gateway capabilities, the proposal expands to include the orchestration, observability, and governance of the agents that take part in these processes. This way, it is not only about providing information to agents, but also about tracking how they use that information and how their actions relate to the rest of the operation.

The opportunity lies in creating processes in which people, automations, and agents can work in a coordinated way, with common context and levels of autonomy compatible with their responsibilities.

Get to know ARGUS and see how it works in practice.

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.

References

[1] McKinsey & Company. The state of AI: How organizations are rewiring to capture value. 2025.

[2] McKinsey & Company. The operating model advantage: Why AI winners are rewiring their organizations. 2026.

[3] Deloitte. Tech Trends 2026.

[4] Deloitte. AI agents are only the beginning: The path to agentic transformation. 2026.

[5] McKinsey & Company. From adoption to impact: Three horizons of AI transformation. 2026.