More than half of companies (57%) expect that, by 2028, artificial intelligence (AI) will autonomously make or execute more than a quarter of their warehouse operational decisions. Furthermore, 90% plan to increase their investment in these technologies over the next two or three years. These are some of the conclusions of the report State of generative and agentic AI in warehouse operations, prepared by the Intelligent Logistics Systems Lab of the Center for Transportation & Logistics (CTL) of the Massachusetts Institute of Technology (MIT) in collaboration with Mecalux.

The Mecalux study, based on the responses of nearly 2,000 professionals specialized in supply chains and warehouse operations from 21 countries, analyzes the degree of implementation of generative and agentic AI, its main applications and the results that organizations are obtaining. It also examines how companies are managing the autonomy of these technologies and human oversight mechanisms as their use in logistics processes expands.

The projects are in their early phases

The level of adoption is already considerable. 61% of organizations say they use generative AI (genAI) at scale or have it fully integrated into their operations, while the percentage stands at 56% for agentic AI. However, many of the projects are still in their early stages: approximately three out of every four applications have been in their current implementation phase for less than a year.

The results point to widespread improvements, although there is still room to measure and document them. Among the use cases described by participants, 87% of genAI applications and 81% of agentic AI applications have achieved at least a moderate improvement in their performance indicators. However, only 39% and 36%, respectively, have recorded significant progress supported by quantified and documented data. In those cases in which respondents were able to estimate the operational impact, the average improvement was around 20% in the main indicator analyzed.

AI agents gain autonomy, but with limits

Companies are beginning to delegate decisions to AI agents, although they maintain restrictions on the operations they can execute without human intervention. Among the functions with greater autonomy, inventory movements stand out, present in 51% of agentic AI use cases, and data corrections, which reach 42%. On the contrary, decisions that involve commitments to clients (22%) or financial issues (12%) remain much more limited.

The most widespread supervision model is the so-called human-on-the-loop. In this approach, AI works within pre-established parameters, while employees monitor its activity and retain the ability to intervene when necessary.

“The goal is to go beyond perceptions and provide practical guidance on the areas where these technologies already provide operational value and those where they are still in the experimental phase,” explains Matthias Winkenbach, director of the Intelligent Logistics Systems Lab.

The benefits of strong governance

Governance and technological integration are emerging as two determining factors for companies to obtain measurable results from their AI projects. According to the report, organizations that have a comprehensive governance model in place and actively apply it are almost twice as likely to achieve significant, documented improvements.

Specifically, this difference is reflected in 51% versus 26% of generative AI cases and 46% versus 23% of agentic AI projects. The trend is maintained even when considering variables such as the degree of implementation, the region, the sector of activity and the available budget. However, the study warns that these data show a relationship between both factors, but do not allow establishing a direct cause and effect relationship.

Technological integration continues to be one of the main obstacles to extending the use of generative AI. 40% of respondents identify the connection with legacy systems as the main barrier to expanding their implementation. Furthermore, 54% consider that better integration with warehouse management systems (WMS) and ERP platforms would be the factor that would most contribute to accelerating their adoption.

More investment and autonomy, along with new professional profiles

Investment forecasts reflect companies’ confidence in the potential of these technologies. Nearly a third of organizations (29%) expect to increase their AI budget by more than 25% over the next two to three years. Overall, 90% plan to increase this item, a figure similar to that recorded in last year’s study, when it reached 87%.

Autonomy will also continue to grow as projects mature. Among companies that have already fully integrated these technologies, 70% expect that, by 2028, agentic AI will autonomously make or execute more than a quarter of their warehouse operational decisions. This forecast exceeds the average of all the organizations analyzed (57%) by 13 percentage points.

In the workplace, the report points to a transformation of professional functions rather than a generalized replacement of jobs. 56% of companies say that AI has allowed their employees to focus on higher value-added tasks. For their part, 11% acknowledge having reduced their workforce, while 9% indicate that they have increased it.

The adaptation of equipment is also gaining prominence. 73% of organizations are already training warehouse workers to acquire new skills, while 61% are developing new technical profiles to respond to the needs posed by the incorporation of these tools into logistics operations.

For those companies that are still defining how to approach this transformation, the report recommends establishing a clear governance framework before expanding AI deployment, placing data quality and systems integration among priorities from the beginning, and evaluating the results of each application during its first 12 to 24 months before deciding whether to extend or discontinue it.

“Warehouse operators need to be able to measure AI performance. This study offers our clients a clear vision of where this technology is working today and what is needed to demonstrate its results,” concludes Javier Carrillo, CEO of Mecalux.