Companies expand AI Agent use, but full automation lags

April 22, 2026 3 min read

Companies are advancing in trials with autonomous systems, yet most are still unable to effectively integrate them into their core processes.

Table of Contents

Are artificial intelligence systems truly transforming companies, or are they still in an extended phase of experimentation?

The answer lies not in the technology itself, but in how organizations are attempting — and struggling — to implement it.

By: Gastón Milano, CTO of Globant Enterprise AI.

AI agents are rapidly advancing within organizations, yet most are still unable to translate this adoption into sustained results.
Reports from Gartner and MIT highlight a persistent gap between experimentation and effective implementation.

It may feel like a much longer journey, but it has been just over three years since the world became familiar with generative artificial intelligence, when ChatGPT was launched in November 2022. That marked only the beginning of a disruptive technology that no organization wanted to ignore once its transformative business potential became evident.

Related Posts:

The result was a massive and accelerated wave of adoption, including the next generation of the technology: AI Agents.

According to a Gartner study, in 2025, 75% of companies experimented with agents, but only 15% implemented fully autonomous systems. The vast majority relied on LLMs combined with automation for highly specific, routine tasks, without fully leveraging their transformative potential.

The diagnosis was similar in the case of GenAI, according to a MIT report. Many companies adopted it simply to avoid missing the hype cycle, and as a result, 95% of pilots failed. So, is the technology not as disruptive as initially promised?

AI adoption and pilots.

Gastón Milano, CTO of Globant Enterprise AI.

What happened in 2025 was the beta phase of AI Agents. The issue was not technological, but organizational. Many companies lacked the workflows required to implement them properly or did not have the human capital prepared to work with them.

The MIT report defines this phenomenon as a learning gap and highlights that any technology, no matter how powerful, requires an adaptation process when integrated into new systems. People play a key role in this process: they must become accustomed to working alongside machines, but above all, they must supervise them and maintain a strategic perspective.

This is one of the keys to success with AI Agents. A recent study — presented as innovative, but in reality confirming what the tech industry already knew — shows that AI lacks context. The challenge is to create it.

Key organizational failures.

One of the most common cases of poor implementation occurs in customer service. With poorly designed systems, users become frustrated and demand human assistance. The issue is not technological, but one of design.

However, there are also organizations that have already learned to leverage AI Agents to lead their industries. A logistics company scaled its support operations, reducing response times from two hours to just 90 seconds. Another example is a semiconductor company that developed an AI Agent capable of solving problems three times faster, with a 75% success rate.

These examples highlight another key lesson in this new phase of AI Agents: they should not be implemented generically, but rather designed with surgical precision to address specific frictions. The real winners will be those who define measurable ROI through focused implementations that deliver real value.

In the next 12 months, 42% of companies plan to develop AI Agents, according to a Gartner report. Adoption is no longer enough: the challenge now is implementation.

Scaling and ROI.

The first step is identifying high-impact solutions to eliminate friction. Next, data must be cleaned and integrated so that AI Agents — designed for specific functions — can operate effectively.

Once orchestration of the agentic system is achieved, along with compliance requirements, it can be scaled to new functions. This process enables a new operating model built on the imperative of measurable ROI. The system must be capable of learning and evolving, rather than being limited to fixed tasks.

This takes on different meanings depending on the industry. In retail, AI Agents build intelligent ecosystems in warehouses to analyze workflows and identify operational bottlenecks; in e-commerce, they can deliver dynamic promotions to reduce cart abandonment; in financial services, they optimize credit decisions and fight fraud in real time. All of this is done autonomously and can be applied across sectors.

The tools already exist and are available. Therefore, the debate is no longer technological, but organizational: what costs are reduced or how much service improves through implementation. The answer to that question will define the success of AI Agents in the coming months.