Three years ago, ChatGPT arrived with the ability to draft a convincing email. Today, the conversation has moved to AI agents: systems that follow multi-step instructions, pull data from several sources, and perform tasks with limited supervision. The technology has outpaced most organizations’ capacity to absorb it.
McKinsey’s November 2025 State of AI report, covering nearly 2,000 organizations across 105 countries, makes the scoreboard plain. 88% of companies use AI somewhere; only about a third have scaled it. Most remain stuck in pilots that never graduate to production. The firms pulling ahead treat AI as an operating change rather than a software purchase: they invest more, redesign how work gets done, and align leadership around it.
A March 2026 working paper from MIT’s Center for Collective Intelligence, by Malone and colleagues, maps where AI creates value across nearly 40,000 work activities, or tasks. Two findings matter: use is extraordinarily concentrated, with 92% of current AI applications clustering on just 6.8% of activities; and information-based tasks absorb roughly 72% of that value, physical tasks only 12%. For shipping, an industry that runs on both, that asymmetry is the shape of the opportunity and the gap.
The gap is already visible
Two ship managers in Piraeus have identical AI subscriptions. In one office, AI drafts routine emails. In the other, it reads charterer vetting questionnaires, cross-references vessel documentation, and drafts responses with citations. Same license, same cost. Different company in two years.
This differs from the software rollouts shipping knows well. ERP, planned maintenance, chartering platforms: these were top-down projects. You bought a system, trained staff, enforced compliance. AI does not work that way. The value lives in how each person uses it on their specific tasks, which means the useful investment is in people, not licenses. The solution cannot be prescribed from above.
Higher-value work is also where the stakes of getting it wrong are greater. Compare three supplier quotations in different formats and currencies: AI does it in minutes and highlights the cheapest option. The trap is that it can confidently misread an Incoterm or miss a mobilization cost in a footnote. The output looks identical to a correct one. Three kinds of people will meet that output. The first takes it at face value and has made a bad decision with the fluency of a good one. The second reads it carefully, catches the error, concludes the technology does not work, and steps back. The third catches the same error, points it out to the AI, and continues the conversation until the answer holds up. Only the third has done what the technology requires. The first two outcomes are failure modes. The third is the system working as designed, with a capable human who knows the material well enough to supervise it. Organizations that cultivate that judgment see it compound. Those that do not keep encountering errors they cannot explain, or abandon tools that would have served them well.
Where the value actually is
Every role in a shipping organization, ashore and at sea, is a bundle of tasks. AI does not replace roles, at least not yet. What it does is replace the bottom 70 to 80% of what each role does, the routine language-based work that fills most of the day. That is a bigger organizational change than replacing the role outright, and most companies are not treating it as one. A port agent email processed in two minutes instead of fifteen. An 18-page drydock report turned into a management summary in three minutes instead of an hour. Three supplier quotations compared in ten minutes instead of two hours. The time saved on the routine 80% is what lets the remaining 20% actually get done well.
What building capability looks like
Leadership alignment comes first. A working session in which leadership uses AI on the company’s own tasks and sees where it succeeds and fails does more than any vendor presentation. Skills come before solutions. The people closest to the friction (the operations coordinator, the superintendent, the crewing officer) know where the pain is; given a working method and practice on their own real documents, they solve their own highest-value problems. That knowledge stays in the company, independent of any platform.
An ERP rollout locks in a system you live with for a decade, and the skill it builds is knowing where the right screen lives and which field to fill. A capability program builds something different: fluency in telling a machine, in plain language, what you need from the documents and data in front of you. That fluency transfers to whatever tool comes next, because the interface is the language your people already speak every day. I have spent the last two years watching Greek shipping companies try to absorb this technology. The pattern is consistent. The companies that buy the biggest platforms are not the ones pulling ahead. The ones pulling ahead pick a small team, pick the right tasks, and teach those people to get the work done with the tools already on their desks. Once that foundation is in place, the next layer opens: AI assistants that pre-draft TMSA self-assessments, respond to vetting questionnaires with citations, turn ERP data into decisions. But only once the foundation is in place. That is what I mean by the capability gap. It is a people problem, not a technology problem, and the companies treating it as a technology problem are the ones falling behind.
I have seen one Greek ship manager take this seriously. They did not buy a bigger platform. A core group of five people, inside an office of fifty, picked a handful of high-friction tasks their colleagues wanted solved and worked on them with real documents from their own operation. They are now pulling data out of the ERP and turning it into something far more actionable than the standard reports it was designed to produce. They are looking at the next layer from the inside out.
What this is actually about
Greek shipping’s position in global markets has always been built on people making sharp decisions with imperfect information, in environments that move faster than the documents describing them.
That is exactly what this technology amplifies, when it is used well. The capability gap is real, it is widening, and it is entirely closable.
Ask one question in your next leadership meeting: which five tasks, across which five roles, would change your operation if AI handled them well? If no one can answer in ten minutes, the problem is not the technology.
Vassilis Papakonstantinou is a partner and the Head of Technology at Blue Dome Capital Limited, P Square Ventures.
