The Exception Was the Whole Year
Co-Founder & CEO of CLRT
If you run logistics, a port or a supply chain in the Gulf, you have spent two years buying operations software with the word intelligent in its name, and you have just lived through a half-year that none of it was built for. This week the two UAE port operators reported. DP World's interim results, filed with Nasdaq Dubai on 13 August, record that Jebel Ali handled 90.1 percent fewer containers in the second quarter than a year earlier. AD Ports Group's release the following day reports UAE container throughput down 65 percent and its strongest quarter on record, on the same page. Read the two documents together and a question the sector has avoided becomes unavoidable: where were the agents while the year was being decided?
Start with the filing. Jebel Ali's gross throughput fell 30.5 percent year on year in the first quarter of 2026 and 90.1 percent in the second, to 374 thousand TEU; across the half it was down 59.5 percent. The rest of the network grew. Excluding Jebel Ali, DP World's gross volumes rose 5.4 percent and revenue rose 18.5 percent, which is why group revenue still climbed 13.1 percent to $12.7 billion while adjusted EBITDA slipped 5.6 percent. The port itself was not the problem. The filing says Jebel Ali remained fully operational with no material physical damage; what closed was the Strait of Hormuz, and the notes date the conflict behind it to 28 February. The filing even introduces a new metric, results excluding Jebel Ali Port, so that readers can see the business through the exception. That is a company saying, in accounting language, that one event outside its yard was the year.
Abu Dhabi's operator tells the same story from the other side of the ledger. AD Ports Group's second-quarter release, dated 14 August, reports UAE container throughput down 65 percent to 573 thousand TEU, attributed to the disruption in the Strait of Hormuz. On the same page, revenue rose 47 percent to AED 7.08 billion, EBITDA rose 49 percent and net profit rose 88 percent to AED 836 million, with asset sales contributing AED 650 million of the revenue. What produced the quarter is described in plain operational prose: cargo and feeder services rerouted from March to Fujairah Terminals and Khor Fakkan, 27 of the group's container vessels and five bulk vessels redeployed onto alternative corridors, bonded overland transit into Khalifa Port and Jebel Ali with 400 additional trucks, six chartered aircraft for food and pharmaceuticals, and warehousing and storage capacity expanded to more than 54,000 square metres. Every item on that list was a decision made by a person.
Now put beside those filings what the same operators publish about AI. Seven weeks before its results, on 23 June, AD Ports launched IHQ, an intelligence headquarters described as employing thousands of digital workers across 20 global workstreams, with demonstrations in port and berth optimisation, vessel arrival orchestration, software development and talent acquisition. Its chief digital officer said every selected workstream sits where AI can measurably improve speed, accuracy and decision quality. The release contains no measurement. The quarterly results that followed do not mention IHQ, digital workers or agents. DP World's half-year filing gives artificial intelligence one sentence: the company is embedding AI and advanced analytics across its operations to optimise planning and improve asset utilisation. No metric is attached. Two operators, one conflict, two sets of reported numbers, and the AI programmes appear in neither set except as a sentence and a headcount of software.
This is not an argument that the agents did nothing. It is an observation about where they were pointed. A berth optimiser is built to squeeze minutes out of a normal day: the vessel arrives, the yard is arranged, the cranes are sequenced, and the gain is measured against yesterday. Its whole value rests on tomorrow resembling the days it learned from. In the second quarter of 2026 tomorrow did not arrive. The vessels did not come through the strait, and Jebel Ali, fully operational, handled a tenth of its usual volume. Meanwhile the work that decided the year, chartering aircraft, negotiating bonded transit, choosing which 27 ships to send where, was exception work, done from relationships, judgment and telephone calls, in a regime no model had seen. The published AI story and the published financial result do not overlap because they describe different halves of the same year.
For an executive who has bought AI-optimised operations, the lesson is uncomfortable but useful. The value of the programme is not the count of digital workers it employs, nor the minutes it shaves off a normal berth. It is what the system does on the day the normal day stops, and the honest answer for most operators is that it is switched off, bypassed or ignored, because it was never pointed at the exception path. The ports that came through the first half of 2026 did so on human improvisation, and that improvisation left almost no published trace: no record in either filing of how the rerouting decisions were made, no log of which trade-offs were made under which information. The sector has automated the part of the operation that was fine and left unmodelled the part that produced the result. Where to point AI next is not the yard. It is the exception, and that is a judgment problem before it is an engineering one.
The sector automated the part of the operation that was fine and left unmodelled the part that produced the result.
A deeper dive
The mechanism deserves precision, because it will recur. An optimiser, whether it schedules berths or sequences cranes, is a function fitted to a distribution: the arrivals, dwell times and yard states it has observed. Its confidence is highest exactly where the data is densest, which is the normal day, and it has no representation of a regime it has never seen, which is a closed strait. When the regime shifts, the system does not fail loudly. It keeps producing well-formed plans for vessels that are not coming, and the people who run the port stop reading them. Nobody logs that moment, so the AI programme's own dashboard shows nothing wrong: the agents ran, the workflows completed, the counts held. DP World's decision to publish an excluding Jebel Ali metric is the accounting version of the same act; the company had to construct a new lens to see its business through the exception, because the ordinary lens could not. An operator's AI programme needs the equivalent, a defined view of what its systems did, and were told to stop doing, during the period that decided the result, and almost none has one.
The second-order traps are what happens next. The first is to read the recovered numbers as validation. Revenue rose, profit rose, the digital workers were deployed, therefore the programme worked; the logic is seductive and the filings do not support it, because the operational account of the quarter is trucks, ships, aircraft and warehouses, and the AI account is a sentence. The second is subtler. Having lived through an exception, the operator now wants to point AI at exceptions, and commissions a resilience agent trained on the data from the first half of 2026. But the knowledge that carried the quarter was largely not in the data: which port authority would take a diverted call, which charterer had capacity, which customs regime would allow bonded transit, which customer needed the pharmaceuticals first. It lived in people, and the next exception will not be this one. The useful move is neither to trust the optimiser nor to build its opposite. It is to decide, function by function, which decisions in the exception path can be structured, evidenced and verified, and which must stay human with a system beside them that keeps the record. That decision cannot be announced. It has to be made.
Work with CLRT
CLRT works on exactly this question, from Dubai, inside the market these filings describe. We do not sell digital workers by the thousand. We sit with an operator, separate the normal-day work an optimiser can carry from the exception work that decided the year, and then engineer the record, the verification and the escalation that let a system earn its place beside the people who handle the exception. If your AI programme cannot tell you what it did in the quarter that mattered, the CLRT Ascent diagnostic at ascent.clrtstudio.com is where that conversation begins, and it begins with where to point it, not with what to buy.
Vishal Sachar is the Co-Founder and CEO of CLRT, where he helps UAE businesses make sense of applied agentic AI and put it to work. He writes on agentic systems, AI governance, and the economics of automation. Reach him at vishal@clrtstudio.com or on LinkedIn.
