AI is poised to transform shipping, an industry that moves about 90 percent of the world’s goods: roughly 11 billion tons of cargo carried by a global fleet of around 110,000 merchant vessels. The opportunity is enormous, but so are the design, operational, and human challenges involved.
Despite the magnitude of this industry, shipping relies heavily on analog measurements and manual processes and, often, suboptimal operational habits. On the other hand, the last few years there is an explosion of technology in this space: cheap sensors, satellite connectivity, onboard computing, and new mathematical models that makes it possible, in principle, to integrate data from engines, weather, fuel use, hull condition, routing, port congestion, cargo, and even crew wellbeing.
Clearly, the discussion and desire are focusing on increased automation and global optimization of ships and fleets. Trying to optimize everything requires access to information essentially “across the board”, which is probably not realistic, at least in the near future, since it is often the case that the competitive advantage is exactly a good piece of information! So, what can this new AI era offer to shipping? A low hanging fruit is optimization of performance and operation at smaller scales: this can range from best strategies in the engine room, to optimal protocols for hull cleaning, and evidence-based decisions for retrofits and performance improvements. In addition, automation of process and procedures within the shipping company, more effective communication across different departments, and better monitoring of errors, are all short-term opportunities that we will see materialize soon.
What is the role of the human in this new era? All these new tools point toward a future in which ships can monitor their own performance, support decisions in real time, and in some cases act with partial autonomy. But, in no way, this future can count on removing the human from the loop. On the contrary, it strengthens human judgment with better information, better models, and better tools. If AI is to succeed in shipping, it must support human decision-making while operating within clear technical, ethical, and operational limits.
The ship’s cognition: Shipping autonomy and agentic AI
The penetration of AI into shipping culminates in the probable creation of autonomous or even partially autonomous ships, the most ambitious and forward-looking idea in the discussions of onboarding AI in the shipping industry. The concept of shipping automation describes how vessels can collect data, calibrate situations grounded on this data, and make in-the-moment decisions independently of human guidance, acting as sovereign systems of perception. The most extreme example is the fully crewless ships, an image that Homer himself most precociously outlined when the Phaeacian king Alcinous described his potent fleet to Odysseus:
“Our ships know by themselves the minds and intentions of men; they know all the cities and the fertile fields of men, and most swiftly they cross the gulf of the sea, wrapped in mist and cloud, and they have no fear of damage or shipwreck.”
However, fully crewless ocean-crossing vessels remain a long way off, for reasons that have less to do with algorithms and more to do with regulation, liability, port infrastructure, and the simple fact that ships still need humans to handle exceptions that no model has ever seen. But narrower forms of autonomy are progressing quickly. Coastal feeder vessels with reduced crews. Autonomous docking assistance. Collision-avoidance systems that genuinely outperform tired human watchkeepers at 3 a.m. Coordinated movements of multiple vessels in confined waters.
The incorporation of agentic AI into ships will be another factor that will allow them not only to anticipate certain outcomes and make predictions but also, with selective directedness, lay out an entire plan and draw out implications – in terms of cost, fuel needs, human work etc. – for this plan as well. Say that a vessel’s fuel consumption is trending upward. The agentic AI can hypothesize that hull fouling is the cause, schedule an underwater inspection at the next port, request quotes from cleaning contractors, and have a human agent evaluate the plan only when this has been completed.
None of this is science fiction; pilots and trials are running today. The serious question is not whether autonomy arrives but how, in what segments first, and under what rules. This new reality will redefine not only how the ship works but what a ship fundamentally is: an intelligent entity with self-cognizant behavior and autonomous reasoning.
Predictive maintenance and engine health monitoring
One of the largest costs for a ship owner is maintenance. The shipping company needs to balance between managing the risk of an expensive failure and the scheduled protocols for maintenance which, sometimes, may be overconservative. If there is one area where AI offers especially clear near-term value, it is predictive maintenance.
Engine failures and auxiliary breakdowns are expensive, disruptive, and sometimes dangerous. One reason these failures remain hard to manage is that much of maritime maintenance is still schedule-based. Components are inspected or replaced after a fixed number of hours whether they need attention or not. Audits are performed on set timetables. This often leads to unnecessary interventions, avoidable cost, and inefficient use of labor and resources.
Predictive maintenance offers a better approach. Instead of relying only on elapsed time, AI models can estimate the condition of specific components based on vibration signatures, temperature trends, lubrication oil analysis, voltage and current measurements, and historical failure records. This makes maintenance more targeted, more efficient, and better aligned with the actual health of the vessel.
Still, the challenge is not only technical. Data quality and integration remain major barriers. Different equipment vendors use different standards, sensor coverage varies across ships, and maintenance records are often inconsistent or incomplete. Without better data infrastructure, even strong models will have limited value.
Designing the ship itself
AI is also reaching upstream into ship design. Hydrodynamics — the science of how hulls move through water — has long relied on a combination of physical model testing and computational fluid dynamics. Both are powerful, both are expensive, and both limit how many candidate designs an engineer can realistically explore. Machine learning surrogates, trained on physics simulations, can evaluate thousands of hull or propeller variants and coupled energy saving devices in the time a traditional method evaluates one. The result is not the end of human design judgment; it is a vastly wider design space to choose from, and ships that are more efficient, easier to manufacture, and better matched to their actual operating profiles.
Combine that with growing interest in alternative fuels, hybrid propulsion, and even small modular nuclear reactors for commercial shipping, and the design problem becomes genuinely new. AI is one of the few tools capable of exploring it at the speed the decarbonization timeline demands.
What still needs to be solved
None of this is automatic. Data in shipping is fragmented across owners, charterers, classification societies, and equipment makers, and many of those parties have legitimate reasons not to share. Connectivity at sea, though improving rapidly with low-earth-orbit satellite constellations, is still uneven. Cyber-physical security is a serious and growing concern: an AI system that can act on a ship is also a new attack surface. Workforce questions loom large — seafarers and shoreside staff will need new skills, and the industry will need new training pipelines to provide them. And the regulatory environment, sensibly cautious, will need to evolve in step with the technology rather than years behind it.
Yet for an industry of maritime’s scale, age, and strategic importance, the opportunity space is unusually open. Performance analysis, predictive maintenance, decision support, ship design, autonomy, and now agentic coordination across the whole value chain are not incremental gains; together, they represent the most significant operational rethink the sector has seen in a generation. Many of the most useful applications have not yet been built. Many of the companies that will build them do not yet exist.
The slow ship is not about to become a fast one. But for the first time in a long time, it is starting to become a genuinely smart one.
Themistoklis Sapsis is a W. Koch Professor of Marine Technology at the Massachussets Institute of Technology (MIT).
