More than half of organisations expect artificial intelligence (AI) to make or execute over a quarter of their operational warehouse decisions autonomously by 2028, signalling a shift towards increasingly automated management of inventory, logistics and warehouse processes, according to a global study by the Massachusetts Institute of Technology (MIT) and warehouse technology company Mecalux.
The report, The State of Generative and Agentic AI in Warehouse Operations, finds that 57% of organisations expect AI to assume this level of decision-making authority within two years, while 90% anticipate increasing their investment in the technology over the next two to three years. Among organisations that have fully integrated AI into their operations, the proportion expecting autonomous systems to handle more than a quarter of warehouse decisions rises to 70%.
The findings are based on 1,967 valid responses collected between 14 August and 1 October 2026 from senior executives and managers in 21 countries. The research was conducted by the Intelligent Logistics Systems Lab at the MIT Center for Transportation & Logistics (MIT CTL), in partnership with Mecalux, and examines the adoption, operational performance and governance of generative AI (GenAI) and agentic AI in warehousing.
While GenAI produces text, summaries, code and recommendations in response to human prompts, agentic AI can carry out multi-step tasks and make decisions with limited or no human intervention. The distinction is increasingly important in warehouse operations, where AI systems are moving beyond information processing towards direct involvement in operational decisions.
Adoption is already extensive. The survey found that 61% of organisations had deployed GenAI at scale or fully integrated it into their operations, compared with 56% for agentic AI. However, the maturity of these applications remains relatively limited: approximately three-quarters of use cases have been at their current deployment stage for less than a year.
The results suggest that businesses are seeing operational improvements, although the extent of those gains is not always supported by formal measurement. At least moderate improvements were reported in 87% of GenAI applications and 81% of agentic AI applications. Yet significant improvements that had been quantified and documented were recorded in only 39% and 36% of use cases respectively. Where organisations could estimate the gains, the typical improvement was around 20% in the principal performance metric being assessed.
AI agents are gaining decision-making authority primarily in internal warehouse activities rather than transactions involving customers or financial commitments. According to the report, 51% of agentic AI use cases allow systems to make decisions independently on inventory movements, while 42% permit autonomous data corrections. By comparison, only 22% allow AI agents to make customer-facing commitments and 12% authorise them to take financial decisions.
Human supervision remains central to the way companies manage these systems. The most common approach is described as a human-on-the-loop model, in which AI operates within established limits while employees monitor its decisions and retain the ability to intervene. This arrangement allows organisations to delegate operational responsibilities without transferring unrestricted authority to automated systems.
Matthias Winkenbach, director of the Intelligent Logistics Systems Lab, said the research was intended to provide practical evidence of the technology’s performance rather than rely on general expectations about AI. “The objective is to move beyond general sentiment and to provide practical guidance on where these technologies are delivering operational value today and where they remain experimental,” he said.
The study also identifies a strong association between effective AI governance and the ability of organisations to demonstrate measurable results. Companies with comprehensive governance frameworks that are actively enforced are approximately twice as likely to report significant, documented improvements as those without such arrangements. For GenAI, the figures are 51% against 26%, while for agentic AI they are 46% against 23%.
The researchers found that this relationship remained evident after accounting for deployment status, geographical region, industry and investment budget. However, the report cautions that the findings demonstrate an association rather than establishing that stronger governance directly causes better performance.
Integration with existing technology remains a major obstacle to wider adoption. Forty per cent of respondents identified the integration of legacy systems as the principal constraint on scaling GenAI, while 54% said improved integration with warehouse management and enterprise resource planning (ERP) systems would have the greatest impact on accelerating adoption.
Investment plans indicate that companies intend to address these challenges while expanding their use of AI. Although 90% expect to increase spending over the next two to three years, 29% anticipate raising their investment by more than 25%. The proportion planning an increase is broadly consistent with the previous year’s study, when 87% reported similar expectations.
The report also examines the effect of automation on warehouse employment. Rather than widespread workforce reductions, the findings indicate that AI is primarily changing the nature of existing jobs. Fifty-six per cent of organisations said the technology had mainly shifted employees towards higher-value tasks. Eleven per cent reported reduced headcount, while 9% said their workforce had increased. To support the transition, 73% are providing additional skills training for warehouse-floor employees and 61% are creating new technical positions.
The report recommends that companies establish clear governance arrangements before expanding AI deployment, give priority to data quality and integration with existing systems, and assess operational results over the first 12 to 24 months before deciding whether to expand or discontinue individual applications.
Javier Carrillo, chief executive of Mecalux, said: “Warehouse decision-makers need AI they can measure. This research gives our clients a clear map of where the technology is working today and what it takes to prove it.”
The study was produced by the Intelligent Logistics Systems Lab at MIT CTL and Mecalux. MIT CTL has more than 50 years of experience in supply chain management research and education, while Mecalux develops automated storage solutions, warehouse management software and metal racking systems. The company has 12 manufacturing plants, seven research and development centres and a workforce of more than 5,500 employees.

