Who’s really in charge when AI runs your ships?

AI-powered systems that not only predict but also decide and execute are rapidly entering shipping operations. A recent MIT Sloan Management Review–BCG study on the emerging agentic enterprise finds that agentic AI is already in use at a significant share of companies, with adoption racing ahead of strategy and governance. Software is starting to behave like a colleague while remaining an asset on the balance sheet—a paradox worth discussing.

Imagine a world where AI agents plan sailings, reallocate boxes, and organise hinterland transport. When a model prioritises speed over safety, or cost over carbon, who carries the can – the vendor, the IT department, the port authority that supplied the data, or the executive who signed off the AI design?

Three schools of thought are emerging. The first places AI in the “supertool” camp. Algorithms crunch data, sense disruption, recommend options and automate standard processes, while humans set objectives, interpret trade-offs and sign off on moves. This is about people and processes, with digital tools amplifying human judgment.

The second school leans into the “digital coworker” language, while wrapping it in governance. Here, AI is considered a teammate, with a role, KPIs and performance reviews. Yet every agent is tied back to a named owner, with rules on what the AI agent can touch, what must be escalated, and how its actions are monitored – an emerging “HR for agents” function that assigns decision rights, guardrails and accountability.​

Rotterdam shows how this looks in practice. With 30,000 seagoing vessels a year, the Dutch port has deployed AI-enabled applications such as Pronto and PortXchange to predict arrivals and coordinate port calls, cutting ship waiting times by 20%. AI-driven scheduling and yard optimisation help match ship arrivals, berth slots, crane allocation and yard capacity, while responsibility for safety, commercial and liability exposure remains with the port authorities, terminal operators and shipping lines.​

The third camp treats agentic AI as a trigger to redraw the operating model. Rather than placing algorithms over legacy programmes and processes, they start from a blank sheet: agents handle fleet, network, and inland logistics rebalancing, while humans focus on resilience, relationships and negotiations. Jobs, incentives and skills are redesigned around human judgment and stewardship at the top of a stack of increasingly capable machines, supported by cross-functional governance that spans shippers, carriers, ports, vendors and regulators.​

However autonomous AI becomes, it does not become a moral or legal agent. Recent maritime studies and legal analyses on automation highlight grey zones around liability as semiautonomous and remote operations expand. Regulators, class societies, and ethicists tend to view AI systems as likely to steer more decisions, but ships still require “seaworthy” human oversight, and humans remain accountable when harm occurs.​​

However, why should we continue with old concepts when software can outplan humans, spot risks earlier and iron out inefficiencies? Insisting on human signoff can look like defending old hierarchies when agentic AI promises better decisions and performance.

Nevertheless, as AI embeds deeper into operations, choices about models, data and guardrails must be owned by leadership and boards, not left to vendors or ad hoc project teams.

Let AI act like a coworker, orchestrating flows at machine speed, while insisting it remains a tool. That means clear ownership for each critical agent, governance that defines escalation procedures and override rights, and a culture that treats every AI decision as a previous human choice. As AI increasingly runs ships and ports, one question will grow louder: when the system acts, who will stand up and say, “I am in charge”?