The conversation around AI agents is changing rapidly. What started with assistants that could answer questions is evolving into systems that can interact with data, use tools and perform tasks on behalf of employees.

According to research referenced by CIO, 75% of organizations already use agentic AI to some degree. Within the next two years, 95% expect to use AI agents. More interestingly, some organizations are moving beyond adoption: 30% are already redesigning key processes around AI, while 34% are starting to use AI for deeper business transformation.

That distinction is important. Deploying an AI agent is relatively easy. Redesigning an organization around what people and AI can each do best is a very different challenge.

The examples in the CIO article illustrate this shift. AT&T has more than 1,000 agentic workflows either in development or already in production. Yet the company is deliberately trying to limit the number of individual agents by making them more flexible and reusable.

The technology itself is also not considered the primary source of differentiation. AT&T has built a model-agnostic platform, allowing models and frameworks to be replaced as the market develops. The company sees its own data, knowledge and expertise as the elements that ultimately create value.

That is an important lesson for organizations currently developing an AI strategy. Building an increasing number of agents does not necessarily create an increasing amount of business value. The question should always be what problem an agent solves, which process it improves and whether AI is actually the appropriate solution.

New York Life provides another interesting perspective. Instead of simply adding AI functionality to individual applications, the company is exploring an AI-powered working environment based on employee roles and objectives. Rather than employees navigating multiple systems themselves, AI can help prepare and orchestrate work across those systems.

This changes the discussion fundamentally. The question is no longer only which tasks can be automated. Organizations need to reconsider how work itself should be designed when AI can increasingly gather information, coordinate activities and execute parts of a process.

For Caronne, this is where the transition to agentic AI becomes primarily an organizational challenge rather than a technological one.

Which decisions may an agent make independently? Which activities still require human judgement? Who remains accountable for the outcome? How do we measure the value created? And where does automation actually make work better rather than simply faster?

Governance is an essential part of that transition. As New York Life emphasizes in the CIO article, the organization remains responsible for what AI does. Delegating work to an AI agent does not mean delegating accountability.

The financial dimension is becoming equally important. Agentic workflows can consume significantly more computing resources because agents repeatedly interact with models and tools. Organizations therefore need to look beyond technical capability and consider whether the business outcome justifies the cost.

This requires a different approach to AI transformation. Rather than asking how many AI agents can be deployed, organizations should identify where autonomy creates measurable value and then design the process, data, controls and responsibilities around that outcome.

The next phase of AI will therefore not be defined by organizations with the most agents. It will be defined by organizations that understand how to combine human expertise and AI effectively.

Technology may enable that transition, but organization, governance and business value will ultimately determine whether it succeeds.


(CIO, artikel, 2026-09-17)