Agentic AI to drive business decisions by 2028

April 21, 2026 4 min read

Between 10% and 20% of processes could be handled by intelligent agents within a few years, with measurable impacts on productivity and work organization.

Table of Contents

Automation is no longer just a promise: intelligent agents are beginning to run business processes.

The challenge now is who supervises them—and under what rules.

Digital transformation is no longer seen as an abstract concept, but as a reality that is beginning to reshape how people work within organizations. In Colombia and Mexico, agentic AI—capable of executing tasks, coordinating processes, and making operational decisions—is starting to be integrated into day-to-day operations, creating opportunities while also introducing new risks that require stronger human oversight.

According to estimates cited by experts, between 10% and 20% of business processes could be managed partially or semi-autonomously by intelligent agents over the next three to five years.
David Alonso, Director of the Technology Department at UDIT, explains:

“We are not talking about full replacement, but about agents capable of executing tasks, coordinating workflows, and making operational decisions in well-defined contexts, always under human supervision.”

Related Posts:

Enterprise adoption.

These projections align with forecasts from firms such as Gartner, which estimates that at least 15% of routine work decisions will be made by agentic AI by 2028, and Deloitte, which anticipates that half of the companies already using generative AI will have deployed agents by 2027.

In an initial phase, adoption will focus on repetitive and structured tasks such as customer service, document generation, information analysis, and software development support. Alonso emphasizes that:

“People will continue to supervise processes with judgment, ensuring accountability and final decision-making within organizations. In this context, the first tasks likely to be delegated to intelligent agents will be the most repetitive, traceable, and standardizable. Rather than full replacement, what we will first see is a reorganization of corporate work, where AI gains weight in execution while humans retain oversight.”

Productivity impact.

The impact on productivity is also beginning to be measured. According to the OECD and studies such as Generative AI at Work and The Impact of AI on Developer Productivity, improvements can range from 5% to more than 25%, depending on the type of task—especially in structured cognitive activities. In more complex processes, the effect is usually more moderate. According to the expert:

“A conservative range would be to expect improvements of 5% to 15% in well-designed complex processes, with some cases exceeding 20%.”

The sectors with the highest level of adoption will be those with digitized operations and large volumes of data, such as financial services and healthcare, according to analyses by McKinsey and the World Economic Forum.

In Latin America, progress continues, albeit unevenly across countries. ECLAC estimates that the region accounts for just 3.7% of global AI demand, reflecting a gap compared to leading economies in research and development. “Colombia and Mexico are better positioned than the average, although they are not yet among the leading group.”

Even so, indicators show sustained growth in enterprise adoption. In Mexico, 89% of leaders plan to incorporate AI agents this year, according to Microsoft’s Work Trend Index.
In Colombia, 90% expect to do so within the next 12 to 18 months. In addition, 43% of Latin American companies already report tangible results from their AI strategies, according to an SAP report.

Risks and oversight.

Beyond the benefits, the deployment of these systems entails operational risks related to errors, cybersecurity, privacy, and loss of control. Alonso warns: “The greatest danger is not that the agent ‘thinks too much,’ but that the organization deploys it without sufficient controls.”

In this context, human oversight becomes central.

“I would recommend at least three levels: an operational level, responsible for supervising the system’s day-to-day execution and able to intervene or stop it if anomalies are detected; a business or process ownership level, responsible for validating metrics, exceptions, and usage criteria; and a third independent level of risk, compliance, or audit, which periodically reviews the system’s performance, incidents, and regulatory compliance. Rather than a single supervisor, a layered and documented oversight approach is more appropriate.”

In this regard, responsibility continues to rest with people: “Responsibility should never be attributed to the system itself, but to a clearly defined human and organizational chain from the outset.”

The advancement of agentic AI also drives the need for risk-based regulatory frameworks aligned with international standards and principles such as transparency, traceability, and accountability, promoted by organizations such as NIST, OECD, and UNESCO.

At the same time, the labor market is beginning to adapt to a hybrid model. Alonso concludes by anticipating that:

“The professional of the future will not be a substitute for AI, but a skilled manager of hybrid work between humans and agents. Skills such as analytical thinking, creativity, and continuous learning will be essential by 2030.”