AI Agents in Logistics: What Is Actually Working in 2026

AI agents are software systems that carry out a whole task rather than a single step. In a logistics business, an agent can read a booking email, match it to a shipment, check whether a slot is free, propose a change and update the record, without a person clicking through each screen. Many logistics companies are buying this technology. Far fewer are running it in production. Gartner surveyed chief supply chain officers and found that 55 percent are unclear on the return from their AI investments, even though 67 percent of supply chain digital spending now goes to AI (Gartner, 5 August 2026). That gap between spending and proof is the real story of 2026, and it is the one worth planning around.

What makes an AI agent different from the automation we already have?

An AI agent adds judgement. It can read messy input such as a free-text email or a scanned document, decide which of several paths to take, call other systems to act, and stop and ask a person when it is not confident. That flexibility is the benefit. It is also the risk, because the agent can be confidently wrong in a way a rule-based script never is.

How many logistics companies actually use AI agents today?

Very few, relative to the noise. The Capgemini Research Institute surveyed 1,000 senior executives at organisations with more than one billion dollars in revenue across 13 countries, with fieldwork in March and April 2025. It found that 9 percent were actually using AI agents or multi-agent systems, while 67 percent expected agentic AI to significantly boost productivity (Capgemini Research Institute, September 2025). Capgemini sells supply chain AI implementation services, so it has a commercial interest in this subject. The 9 percent figure works against its own sales message, which is part of why it is worth quoting.

Two Gartner surveys point the same way. Only 23 percent of supply chain leaders had a formal AI strategy in place (Gartner, 11 June 2025), and only 17 percent were redesigning processes around AI rather than adding it to existing ones (Gartner, 6 May 2026).

Adoption of AI in some form is much broader. The 2026 MHI Annual Industry Report, produced with Deloitte from a survey of 500 supply chain professionals, found 41 percent of companies using AI in their supply chain, up from 30 percent the year before (MHI and Deloitte, 15 April 2026). MHI is an industry association and Deloitte sells transformation consulting, so both have an interest in a rising adoption number. The useful reading is that ordinary AI use is now common, while agents that act on their own are still rare.

Where does AI actually pay off in logistics right now?

In high-volume work that is rule-bound and cheap to check. Appointment scheduling, driver follow-up calls, customs and document data entry, exception alerts, slot booking and first-line customer queries all share the same quality: a wrong answer is caught quickly and costs little to correct.

The largest published deployment fits that pattern. DHL Supply Chain deployed AI agents that handle phone and email interactions for appointment scheduling, driver follow-up and warehouse coordination, at a scale of hundreds of thousands of emails and millions of voice minutes a year (DHL Group, 11 November 2025). It is worth noting what that announcement does not contain. DHL described the result as significantly reducing manual effort and published no percentage figures.

That absence matters. We could not find an independently measured improvement in forecast accuracy from a production logistics deployment anywhere in the public record. Every figure we traced led back either to vendor marketing or to older projections. Treat that number as unavailable rather than assuming it is good.

The labour pressure behind all of this is real. The IRU reports around 502,000 unfilled truck driver positions in Europe, a shortage rate of 13 percent, with roughly 20 percent of the driver workforce due to retire within five years (IRU, 30 June 2026). The IRU is a road transport industry body that campaigns for policy support on exactly this issue, so read the figures with that interest in mind.

What about the statistics everyone quotes?

The claim that AI improves logistics costs by 15 percent, inventory levels by 35 percent and service levels by 65 percent comes from a McKinsey article published on 30 April 2021. The original wording compares early adopters with slower-moving competitors. It is not a before and after measurement of a single company.

The claim of up to 50 percent fewer forecasting errors and 65 percent fewer lost sales comes from a McKinsey report published in April 2017, where it was written as an expectation rather than an observed result. DP World republished that figure as a present-tense finding in September 2025 without citing the underlying study (DP World, 16 September 2025).

Neither number is dishonest in its original form. Both become misleading when a 2026 proposal presents them as current evidence. A simple question protects you: what is the publication date of the study behind this number, and was the result measured or projected?

What goes wrong?

The first is buying something that is not what it claims to be. Gartner estimates that of the thousands of vendors describing themselves as agentic, only around 130 have genuine agentic capability, and it forecasts that more than 40 percent of agentic AI projects will be cancelled by the end of 2027 because of rising costs, unclear business value or weak risk controls (Gartner, 25 June 2025).

The second is integration. When Gartner asked 140 senior supply chain leaders what blocked them from scaling AI, 56 percent named integration with legacy systems and processes and 50 percent named limited internal expertise (Gartner, 29 April 2026). The model is rarely the hard part. The connection to the transport management system usually is.

The third is over-automation. Kroger recorded 2.6 billion dollars in impairment and related charges on its automated fulfilment network (Kroger, 4 December 2025) and agreed to pay Ocado 350 million dollars in compensation after closing three automated centres and cancelling a fourth (Supply Chain Dive, 10 December 2025). That is robotics rather than AI models, and it is an audited public disclosure rather than a survey. The failure mode transfers directly: throughput never reached the level needed to pay for the technology. An agentic AI business case can fail the same way.

What is happening in the UAE and the wider GCC?

Regional activity is running ahead of published evidence, in the same pattern seen everywhere else. AD Ports Group announced the launch of an AI-powered Intelligence Headquarters in June 2026 (AD Ports Group, 23 June 2026). As with the DHL deployment, the announcement describes the capability rather than a measured result, so it is evidence of direction and not of return.

The sector is large enough for small efficiency gains to be worth pursuing. Press reporting in September 2025 put the logistics sector's expected contribution to the UAE economy above AED 200 billion (Gulf Today, 7 September 2025). That figure is a newspaper account of a forward projection rather than a measured outcome, so it belongs in a market sizing slide and not in a business case.

For UAE operators moving goods into or through the European Union, two compliance dates matter more than any model choice. The EU Import Control System 2 extended to rail and road transport in April 2025, and the EU Data Act has applied since 12 September 2025. Both affect what data an operator has to produce and share, and both are exactly the kind of recurring, document-heavy obligation that teams try to automate first.

Where should a logistics business start?

Pick a single high-volume process with a clear definition of done, such as appointment scheduling or customs document preparation. Agree the measure before the build, in hours saved or cycle time cut. Give the agent access only to the data that process needs. Keep a named person approving anything consequential, and log every action so it can be reviewed and reversed. Prove it there before widening the scope.

This is the same discipline set out in our AI agent governance framework and our analysis of where AI automation returns actually appear. It is also the opposite of the cancelled 40 percent, which almost always begin with broad ambition and no agreed measure of success.

Frequently asked questions

Are AI agents ready for logistics operations?

For narrow, high-volume, rule-bound tasks, yes. For open-ended decision making, no. Only 9 percent of surveyed supply chain organisations are actually using AI agents, against 67 percent who expect significant productivity gains from them (Capgemini Research Institute, September 2025), and Gartner expects more than 40 percent of agentic AI projects to be cancelled by the end of 2027 (Gartner, 25 June 2025). Bounded scope and human approval are what separate the two outcomes.

What is the single biggest reason these projects fail?

Integration with existing systems, named by 56 percent of senior supply chain leaders, followed by limited internal expertise at 50 percent (Gartner, 29 April 2026). Cost overruns, unclear business value and weak risk controls are the reasons Gartner gives for project cancellation.

Does the EU AI Act apply to a logistics AI system?

It depends on the use, and the timing has changed. High-risk obligations now apply from 2 December 2027 for stand-alone systems and 2 August 2028 for AI embedded in regulated products (Council of the EU, 29 June 2026). Prohibitions and AI literacy duties already apply. A route optimiser is not automatically high-risk, but AI used in a safety component of machinery or a vehicle may be, so classification should be done early rather than assumed. This is general information rather than legal advice, and the classification of a specific system should be confirmed with counsel.

How should we judge a vendor's performance claims?

Ask for the publication date of the underlying study and whether the result was measured or projected. Two of the most quoted logistics AI statistics date from 2021 and 2017 respectively, and the 2017 figure was written as an expectation rather than a measured outcome.

Elchai Group builds and governs AI agents for logistics and supply chain operations across the GCC and Europe, pairing agent engineering with the permission boundaries, human approval steps and audit trails that keep an automated workflow accountable to a named person.

Sources

• Gartner, Gartner Survey Finds Majority of Chief Supply Chain Officers Unclear on AI Investment Returns, 5 August 2026
• Capgemini Research Institute, New-generation supply chain, September 2025 (fieldwork March to April 2025, 1,000 executives, 13 countries)
• Gartner, Just 23% of Supply Chain Organizations Have a Formal AI Strategy, 11 June 2025
• Gartner, AI Is Not Driving Supply Chain Operating Model Transformation, 6 May 2026
• MHI and Deloitte, 2026 MHI Annual Industry Report: Rewiring the Future, 15 April 2026
• AD Ports Group, AD Ports Group Launches AI-Powered Intelligence Headquarters, 23 June 2026
• Gulf Today, Logistics sector's contribution to UAE's economy to exceed Dhs200 billion, 7 September 2025
• DHL Group, DHL boosts operational efficiency and customer communications with HappyRobot's AI agents, 11 November 2025
• IRU, Operators deeply concerned by worsening driver shortage, 30 June 2026 (2025 data)
• McKinsey & Company, Succeeding in the AI supply-chain revolution, 30 April 2021
• McKinsey & Company, Smartening up with Artificial Intelligence: What's in it for Germany, April 2017, page 33
• DP World, Companies using AI report up to 50 percent reduction in forecasting errors, 16 September 2025
• Gartner, Over 40% of Agentic AI Projects Will Be Canceled by End of 2027, 25 June 2025
• Gartner, Technology Integration and Talent Perceived as Key Roadblocks to Scaling AI in Supply Chain, 29 April 2026
• The Kroger Co., Kroger Reports Third Quarter 2025 Results, 4 December 2025
• Supply Chain Dive, Kroger canceling Charlotte CFC, closing Nashville spoke, 10 December 2025
• Council of the EU, Artificial intelligence: Council gives final green light to simplify and streamline rules, 29 June 2026
• European Commission, Data Act explained (applicable since 12 September 2025)
• European Commission, Taxation and Customs Union, EU Import Control System 2 extends to rail and road transportation in April 2025, 3 February 2025